{
 "version": 1,
 "generatedAt": "2026-11-22T14:07:37Z",
 "site": {
  "name": "QDev",
  "url": "https://quantumdev.io/",
  "llmsTxt": "https://quantumdev.io/llms.txt",
  "sitemap": "https://quantumdev.io/sitemap.xml",
  "feed": "https://quantumdev.io/blog/feed.xml"
 },
 "data": {
  "events": {
   "url": "https://quantumdev.io/events/events.json",
   "schema": "https://quantumdev.io/events/events.schema.json",
   "calendar": "https://quantumdev.io/events/events.ics"
  }
 },
 "licence": {
  "url": "https://quantumdev.io/license.xml",
  "terms": "https://quantumdev.io/terms/#ai"
 },
 "mcp": {
  "url": "https://quantumdev.io/mcp",
  "serverCard": "https://quantumdev.io/mcp/server-card"
 },
 "documents": [
  {
   "id": "page:/",
   "type": "page",
   "title": "The Quantum Environment",
   "url": "https://quantumdev.io/",
   "markdownUrl": "https://quantumdev.io/index.md",
   "description": "Quantum computing for developers. The Quantum Environment spans four products: Quantum Learner, Quantum Architect, Quantum Instructor, and Quantum Shop.",
   "updated": "2026-11-11",
   "text": "# QDev\n\nCanonical: https://quantumdev.io/\n\n## Products\n\nFour products. One is open.\n\n1. ### Learn\n\n   Quantum computing, taught in a developer's terms.\n\n   Coming soon\n2. ### Architect\n\n   For quantum programs of your own.\n\n   Coming soon\n3. ### Instruct\n\n   Tools for teaching the field to the next cohort.\n\n   Coming soon\n4. ### Shop\n\n   Objects and materials worth keeping.\n\n   [shop.quantumdev.io](https://shop.quantumdev.io)\n\n## QDev is one environment for developers who work with quantum computing. Learn it, build with it, teach it, and take a piece of it home.\n\nWe think a working developer should be able to pick up quantum computing the way they pick up any new stack.\n\nEvents\n\n## Quantum events, on a globe you can turn.\n\n[Open Events](https://quantumdev.io/events/)\n\nRecommended reading\n\n## MIT Quantum Index Report 2026\n\n[qir.mit.edu](https://qir.mit.edu/)\n\n## Join the list\n\nOne email when each product opens and when an event is announced.\n"
  },
  {
   "id": "page:/events/",
   "type": "page",
   "title": "Events",
   "url": "https://quantumdev.io/events/",
   "markdownUrl": "https://quantumdev.io/events/index.md",
   "description": "Find quantum computing events by continent, country, state and city, each linked to its organiser.",
   "updated": "2026-11-10",
   "text": "# Events\n\nCanonical: https://quantumdev.io/events/\n\nQuantum events around the world, each run by its own organiser.\n\nThe page is a search by place. A visitor picks a continent, then a country, then a state where the country has them, then a city, and sees the events there, beside a calendar of the months ahead, the events nearest a region they choose, and the other events that fit the same trip. QDev checks each event against its organiser's page and shows the date of that check.\n\n## The events as data\n\n- [events.json](https://quantumdev.io/events/events.json): every published event that is not over yet, soonest first. It is rewritten on every publish and once a day, and its events list is empty when nothing is published.\n- [events.schema.json](https://quantumdev.io/events/events.schema.json): the JSON Schema of that file.\n- [events.ics](https://quantumdev.io/events/events.ics): the same events as a calendar feed to subscribe to.\n- [The MCP server](https://quantumdev.io/mcp): tools to list places and events, look one up and get it as a calendar file.\n\n## Linking to a place or an event\n\n- `https://quantumdev.io/events/?event=<id>` opens one event, where `<id>` is the event's `id` in events.json.\n- `https://quantumdev.io/events/?in=<continent>/<country>/<city>` opens a place. Each step is the place's name in lowercase with hyphens for spaces, such as `europe/germany/munich`. A country that is searched by state has the state between the country and the city.\n"
  },
  {
   "id": "page:/privacy/",
   "type": "page",
   "title": "Privacy Policy",
   "url": "https://quantumdev.io/privacy/",
   "markdownUrl": "https://quantumdev.io/privacy/index.md",
   "description": "What QDev Inc. collects across quantumdev.io and its products, why, who else sees it, and how to reach us about it.",
   "updated": "2026-11-21",
   "text": "Legal\n\n# Privacy Policy\n\nCanonical: https://quantumdev.io/privacy/\n\nUpdated 2026-11-21\n\nQDev Inc. runs quantumdev.io and the products on it. This policy explains what we collect, why we collect it, who else sees it, and how to reach us about it. It covers this site, Quantum Learner, Quantum Architect, Quantum Instructor and Quantum Shop.\n\n## Who we are\n\nQDev Inc. (\"QDev\", \"we\", \"us\") is a Delaware corporation and is responsible for the personal information described here. For anything about your privacy, write to [info@quantumdev.io](mailto:info@quantumdev.io).\n\n## What we collect\n\n**On this site**\n\nquantumdev.io sets no cookies and runs no analytics. Its two forms are the email list sign-up and the member account, both described next; if you use neither, the site collects nothing that identifies you. A few choices are saved in your own browser and never sent to us: light or dark; that you closed the sign-up, kept for 30 days so it stays closed; that you joined, kept until you clear your browser's data, so it doesn't ask again; and the region you choose on the Events page, which that page uses to work out distances. If you sign in to an account, the site also keeps a sign-in token in that browser for up to 30 days, or until you sign out, so you stay signed in on that device. Like any website, our hosting provider receives standard request details, such as your IP address, your browser type and the page you asked for, so it can deliver the page and protect the site from abuse.\n\n**When an AI assistant uses our MCP server**\n\nquantumdev.io has a server that lets AI assistants read our public events, pages and posts, and start an email list sign-up when you ask for one. A sign-up started that way works as the one on this site does: we email you to confirm, and we record that it came through the server. For each request we record the time, the name of what was asked for (such as \"list events\"), the name and version the assistant's software gives for itself, and the size of the answer. We don't record what you typed to your assistant or the details of the request. As with any request to the site, our hosting provider also records the address it came from, which is usually the assistant company's and not yours. We keep these records for 90 days and use them to count how the server is used and to stop abuse.\n\n**When you join the email list**\n\nYour email address, the page you joined from, the time, and the exact wording of the consent you saw. If you answer the optional questions about which of our products interest you, what brings you to quantum computing and which tools you build with, your answers too. If you join from one product's own sign-up control, that product starts chosen and you can change it before you close the sign-up. We use these answers to see who is on the list, what they work with and which products they are waiting for, so we can decide what to build and write about first. Choosing a product does not change which emails you get. We email you to confirm; until you confirm, we don't add you to the list, and we delete an unconfirmed request about 30 days after the address was first entered.\n\n**When you create an account**\n\nYour email address and a password, which we store only in a form that cannot be read back. If you fill in your profile: your name, a headline, your pronouns, what best describes you, your company or school, your city and country, what brings you to quantum computing, the tools, hardware and subjects you work with, what you are working on now, what you are open to, a description of yourself, links to your pages elsewhere, your experience and your education. You can also upload a picture and a banner. Before we store an image we remove the data a camera adds to it, such as where it was taken, and an automated check run by Amazon Web Services looks for content we cannot show. Every part of a profile is optional. Your profile is private until you make it public. If you choose a handle and make it public, anyone can read the parts you have not held back, at quantumdev.io/u/ followed by your handle, together with your member number; you choose whether search engines may list it and whether AI assistants may read it, and we do not allow anyone to train an AI model on it. Making it private again takes the page down within a minute, though we cannot recall copies others made while it was public. A stored picture has a long random address; while your profile is private the site shows that address only to you, and anyone you gave it to could open the picture. If you save an event from the Events page, or mark that you are going, we keep which events and when you saved each one. Only you can see them unless you choose otherwise, and each choice is off until you make it. If you choose to show your events on your profile, anyone who can read your profile can read the events you marked that you are going to and the ones you went to; events you only saved are never shown. If you choose to be listed among who is going, your name, what you are open to and a link to your profile appear on the page of each event you marked that you are going to, where anyone can read them, search engines included. A list appears once three members have joined it, and you are on it only while your profile is public. An event's page is kept for up to two years after the event, and the list stays on it as the members who went. To make the lists we keep, for each event, which members chose to be listed. Unticking either box takes your events or your name down within a few minutes, though we cannot recall copies others made. If you send an event for the list, or a correction to one, we keep what you typed, when you sent it and what we decided, and you can see each of them on the account page. We read it, with an AI agent that helps us check events, before anything is listed; nothing you send is shown to anyone else unless we list it. If we list an event you sent and your profile is public, the event's page names you as the member who added it, and otherwise it says a member added it. You can withdraw a submission while it waits. If you send a link from an event, to slides, a recording or a write-up, we keep it as we keep a submission, and we open it, with the same help, before it shows. An approved link shows on the event's page without your name. If you say it is your own talk and you have chosen to show your events on your profile, your profile lists it under “Spoke at”. You can take a link down on the account page at any time. If you run an event on the list, you can claim its listing. To prove it we email a code to an address you give at the event's own site; we do not keep that address, only a scrambled form of it, deleted within a few days, so that it is not written to twice in one day. Where a code cannot prove it, you tell us why the listing is yours and we decide. We keep that you organise the event and how that was shown, and if your profile is public the event's page names you as its organiser. An organiser sees how many members keep their event and how many are going, as two numbers, and never who. You can give a claim up on the account page. If you ask for a calendar address, we make a long random address that shows those events and their deadlines to any calendar app that opens it, so anyone you give it to can read them; you can replace it or turn it off on the account page, and the old address stops working at once. If you turn on deadline reminders, which are off until you do, we email the address you sign in with about a week before and the day before each dated deadline of the events you keep, and we keep which reminders we have sent so that none is sent twice. The emails are sent by Resend, as our other emails are. You can also ask to be emailed when an event new to the list fits your profile, at most once a day; to do that we compare your city, your country and the areas, tools and hardware on your profile with each new event, and keep which events we have told you of. You can turn either kind of email off on the account page, or both with the link in any of them, and then we send no more. You can download everything your profile holds, your saved events among it, from the account page. If you turn on an authenticator app, we hold the key that links it to your account. An account on quantumdev.io does not ask for a phone number; if a product ever needs one to secure your account, it will say so when it asks. While you use a product, we keep what it needs to work for you, such as your progress.\n\n**When you order from the Shop**\n\nYour name, email address and shipping address, the fit and size you choose, the order amount, and the edition number assigned to you. Stripe handles payment: your card details go to Stripe, and we never see or store them.\n\n**When you join a Shop waitlist**\n\nYour email address and a record that you confirmed it. You confirm by following a link we email you; until you do, we send you nothing else.\n\n**When you write to us**\n\nYour message and whatever details you include, so we can answer.\n\n## How we use it\n\n- To run the products: your account, your progress, and help when you ask for it.\n- To take, make and ship Shop orders, and to email you about them, including your confirmation and edition card.\n- To send the email list's emails you confirmed: one when each product opens and one when an event is announced. Every one has a link to unsubscribe in one click.\n- To send the waitlist emails you signed up for. Every one has a link to unsubscribe.\n- To keep accounts and the products secure, and to prevent fraud and abuse.\n- To measure whether our ads on X lead to Shop sales, as described under Advertising and cookies.\n- To meet our legal, tax and accounting obligations.\n\nWe use your phone number only to secure your account and to contact you about it. We do not send marketing text messages.\n\nWe do not sell your personal information.\n\n## Who else sees it\n\nWe share personal information only with companies that help us run the products, and only what each of them needs:\n\n- **Amazon Web Services** hosts the site and products, stores our data, delivers our emails and checks the pictures members upload.\n- **Resend** keeps the email list's confirmed addresses, with your answers to the optional questions (the products you chose among them), the page you joined from and the version of the consent wording, and delivers the list's emails, including the confirmation.\n- **Stripe** processes Shop payments.\n- **The companies that make and ship your order** receive your name, your shipping address and what you ordered.\n- **X** receives the advertising information described below.\n\nWe may also disclose information when the law requires it, to protect someone's safety or our rights, or as part of a merger, sale or reorganization of QDev. Anyone who takes on your information that way must keep honoring this policy.\n\n## Advertising and cookies\n\nquantumdev.io uses no cookies or trackers, and the email list's emails contain no tracking pixels or tracked links. Products with accounts use the cookies or browser storage needed to keep you signed in.\n\nThe Shop advertises on X (formerly Twitter). If you allow it, Shop pages load X's pixel, which records your visit and sets X's own cookies so we can see how our ads perform. If you decline, the pixel doesn't load. You can change your choice at any time from the Shop's footer.\n\nWhen you buy, we report the sale to X: the amount, the currency, and a hashed (encoded) form of your email address. X can match that hash only to an X account that uses the same address. What X does with this information is governed by [X's privacy policy](https://x.com/en/privacy).\n\nOur products don't respond to browser Do Not Track signals; the Shop's advertising follows the choice you make in the Shop instead.\n\n## What's public\n\nEach Shop edition has a public page, opened by its code, so anyone can verify it. It shows the drop, the edition number and the code. It never shows your name, email or address.\n\n## How long we keep it\n\n- **Account information:** while your account is open. When you close it, we delete it within 30 days, except anything the law requires us to keep.\n- **Shop orders:** as long as tax and accounting law requires. The edition record itself, its drop, number and code, stays without your details so the edition remains verifiable.\n- **Email list:** your address stays on the list until you unsubscribe. When you do, Resend keeps your address marked as unsubscribed so you get no more list emails, and we keep the record of your confirmation (your address, when you confirmed, the wording you agreed to and your optional answers, the products you chose among them) as proof that you agreed, until you ask us to delete it. Unconfirmed requests are deleted about 30 days after the address was first entered. Our database backups, which may hold any of this, are kept for 35 days. Records of requests to the list's server are kept for 90 days and contain no addresses.\n- **Waitlist:** until you unsubscribe.\n- **Request details held by our hosting provider:** a short period, for security and troubleshooting.\n\n## Your choices and rights\n\nYou can ask us to show you the personal information we hold about you, correct it, delete it, limit how we use it, or send you a copy. You can also object to how we use it, or withdraw a permission you gave, such as for the email list, a waitlist or advertising, at any time; withdrawing doesn't affect what we did before. Write to [info@quantumdev.io](mailto:info@quantumdev.io) from the email address you used with us; we may ask you to confirm it's you. We answer within 30 days.\n\n**If you are in the EU or UK:** we rely on our contract with you to run your account and fulfill your orders; on your consent for the email list, waitlist emails and the Shop's advertising pixel; on our legitimate interests in security, fraud prevention and measuring whether our ads lead to sales; and on legal obligations for tax and accounting records. You can also complain to your local data protection authority.\n\n## Where it's stored\n\nWe are based in the United States, and your information is stored and processed there with Amazon Web Services, and the email list's addresses with Resend. If you are outside the United States, your information is transferred there; where the law requires, we rely on safeguards such as the European Commission's standard contractual clauses.\n\n## Security\n\nInformation travels over encrypted connections, access to personal information is limited to the people who need it to run the products, and payment details never reach our systems. No system is perfectly secure. If a breach affects you, we will tell you as the law requires.\n\n## Children\n\nOur products are not meant for children under 13. You must be at least 13 to create an account, or older where your country requires it, and if you are under 18 you need a parent or guardian's permission. If we learn we have collected information from a child under 13, we delete it. A parent who believes this has happened can write to [info@quantumdev.io](mailto:info@quantumdev.io).\n\n## Changes to this policy\n\nIf we change this policy, we'll post the new version here with a new date. If a change significantly affects how we use your information, we'll tell account holders by email before it takes effect.\n\n## Contact\n\nPrivacy questions and requests: [info@quantumdev.io](mailto:info@quantumdev.io).\nHelp with an order or account: [support@quantumdev.io](mailto:support@quantumdev.io).\n\nQDev Inc., a Delaware corporation, 675 Sturdivant Drive, Clarksville, TN 37042.\n"
  },
  {
   "id": "page:/terms/",
   "type": "page",
   "title": "Terms of Use",
   "url": "https://quantumdev.io/terms/",
   "markdownUrl": "https://quantumdev.io/terms/index.md",
   "description": "The terms for using quantumdev.io and its products, including Quantum Shop orders, from QDev Inc.",
   "updated": "2026-11-19",
   "text": "Legal\n\n# Terms of Use\n\nCanonical: https://quantumdev.io/terms/\n\nUpdated 2026-11-19\n\nThese terms are the agreement between you and QDev Inc. for using quantumdev.io and its products, including orders from the Quantum Shop. They're written to be read. If anything is unclear, ask us at [info@quantumdev.io](mailto:info@quantumdev.io).\n\n## The agreement\n\nThese terms are an agreement between you and QDev Inc. (\"QDev\", \"we\", \"us\"), a Delaware corporation. They cover quantumdev.io and the products on it: Quantum Learner, Quantum Architect, Quantum Instructor and Quantum Shop. Using any of them means you accept these terms and our [Privacy Policy](https://quantumdev.io/privacy/). If you don't accept them, please don't use the products.\n\nThe Quantum Shop section adds terms that apply to orders.\n\n## Who can use the products\n\n- You must be at least 13, or older where your country sets a higher age for using online services (up to 16 in parts of the EU).\n- If you are under 18, a parent or guardian must agree to these terms for you.\n- To order from the Shop you must be 18, or the age of majority where you live, or have a parent or guardian place the order.\n- You can't use the products where the law forbids you to.\n\n## Your account\n\nGive accurate details when you sign up and keep them current. Keep your sign-in secure: you're responsible for what happens under your account. If you think someone else is using it, tell us at [support@quantumdev.io](mailto:support@quantumdev.io).\n\nYou can close your account at any time from your account page, which deletes your profile and your sign-in straight away, or by writing to [support@quantumdev.io](mailto:support@quantumdev.io).\n\n## Using the products fairly\n\nWe give you a personal, non-transferable permission to use the products under these terms. Please don't:\n\n- break the law or anyone else's rights;\n- try to get into accounts or systems that aren't yours, or probe, overload or disrupt the products;\n- copy, scrape, resell or republish our lessons or other content, except as the law or the section on AI and automated access allows;\n- reverse engineer the products, except where the law allows it;\n- upload anything harmful, such as malware, or anything you don't have the right to share;\n- pretend to be someone else.\n\nIf you break these terms, we may suspend or close your account. Where it's fair to, we'll tell you why first.\n\n## Learning content\n\nOur lessons and materials are for education. We work to keep them accurate, but quantum computing moves fast, and we can't promise that everything is complete or current or that you'll reach a particular result. Nothing in them is professional advice.\n\n## What you create\n\nAnything you write or build in the products stays yours. You give us permission to store, process and display it as needed to run the products for you, and you're responsible for having the right to share it. We may remove content that breaks these terms.\n\nYour profile is private until you make it public. If you do, you give us permission to show it to anyone at its address on quantumdev.io and, where you allow it, to search engines and AI assistants. You can make it private again or delete it at any time, and that permission ends when you do. Copies that others made while it was public are beyond our reach.\n\nDon't put on a profile anything unlawful, anything you have no right to share, anything that harasses or impersonates someone, or advertising. We may hide a profile that breaks these terms. To report a profile, or to tell us that something on one infringes your copyright, write to [support@quantumdev.io](mailto:support@quantumdev.io) with the profile's address and what is wrong.\n\nIf you send us feedback or ideas, we may use them without owing you anything.\n\n## Our content and name\n\nThe products, their content and software, and the names QDev and The Quantum Environment and the grid-Q mark belong to QDev Inc. or our licensors. These terms don't give you any right to use our names or marks.\n\n## AI and automated access\n\nSoftware may read quantumdev.io. Search engines may index it, and AI assistants may read it to answer a question. Neither needs our permission or a payment.\n\nTraining or fine-tuning an AI model on our content needs a licence from us. Without one, it is not permitted.\n\nA training licence covers every page and post on quantumdev.io and our events catalog. It is for one model developer, lasts three years and is not exclusive. It costs US$0.005 for each word of our pages and posts on the day the licence starts, with a minimum of US$500. To ask for one, write to [info@quantumdev.io](mailto:info@quantumdev.io).\n\nThe events in our catalog are run by their own organisers. A licence covers our record of them, not the organisers' own material.\n\nMembers' profiles are their own. A training licence does not cover them, and a member's public profile may not be used to train an AI model. Each member chooses whether search engines may list their profile and whether AI assistants may read it.\n\nThese terms are also published for software to read, in [license.xml](https://quantumdev.io/license.xml) and [robots.txt](https://quantumdev.io/robots.txt).\n\n## Quantum Shop\n\n**How drops work**\n\nEach drop is a limited edition sold by pre-order during a set window. The Shop page states the object, the price, the edition size and when the window closes.\n\n**Your order**\n\nPrices are in US dollars and shown at checkout, and Stripe processes payment. Your order is accepted when payment completes and we send your confirmation. If we can't fulfill an order, for example because of a pricing error or suspected fraud, we'll cancel it and refund you in full.\n\n**Edition numbers**\n\nEditions are numbered in the order purchases complete, and your number is final. A refunded edition is retired and never issued again.\n\n**Shipping**\n\nWe ship worldwide after the pre-order window closes, and delivery dates are estimates. Check your shipping address at checkout: we can't be responsible for delays or losses caused by an incorrect address. Orders delivered outside the United States may be charged import duties or taxes by the destination country, and those are yours to pay.\n\n**Cancellations and refunds**\n\nYou can cancel for a full refund at any time before the pre-order window closes: write to [support@quantumdev.io](mailto:support@quantumdev.io) from the email address on your order. Once the window closes, your edition goes into production, and it can't be cancelled or changed to another size. If your order arrives damaged, defective or not what you ordered, write to us within 30 days of delivery with a photo, and we'll replace it or refund you. Refunds go back to your original payment method.\n\n**Your legal rights**\n\nNothing here takes away rights that consumer law gives you where you live, including any right to cancel a purchase made at a distance in the EU or UK.\n\n## Other services\n\nThe products link to and rely on services we don't control, such as Stripe, X and LinkedIn. Their own terms apply when you use them, and we aren't responsible for them.\n\n## Changes to the products\n\nWe may change, pause or stop any product or feature. If that affects a Shop order you've paid for, we'll fulfill it or refund you in full.\n\n## Disclaimer\n\nTo the extent the law allows, the products are provided as they are and as available, without warranties of any kind, express or implied, including that they will be uninterrupted, error-free or fit for a particular purpose. Goods you buy from the Shop keep any warranties that consumer law gives you.\n\n## Limits on liability\n\nTo the extent the law allows, QDev is not liable for indirect, incidental, special or consequential losses, or for lost profits or data, arising from the products or these terms. Our total liability for any claim is limited to the greater of what you paid us in the 12 months before the claim or US$100. These limits don't apply where the law doesn't allow them, such as for fraud, or for death or personal injury caused by negligence.\n\n## If you misuse the products\n\nIf you break these terms or the law while using the products, and someone brings a claim against QDev because of it, you agree to cover our reasonable costs of dealing with that claim, to the extent the law allows.\n\n## Ending\n\nYou can stop using the products at any time. We may suspend or end your access if you break these terms. Sections that by their nature should continue, such as those on our content, liability and disputes, still apply afterwards.\n\n## Disputes and governing law\n\nIf you have a problem, write to [info@quantumdev.io](mailto:info@quantumdev.io) first and give us 30 days to try to put it right. These terms are governed by the laws of the State of Delaware, without regard to its conflict-of-laws rules, and disputes will be heard in the state or federal courts located in Delaware. If you are a consumer living outside the United States, you also keep the protection of your own country's mandatory laws and can bring claims in your local courts.\n\n## Changes to these terms\n\nWe may update these terms. We'll post the new version here with a new date, and for significant changes we'll tell account holders by email before they take effect. Using the products after that means you accept the updated terms.\n\n## General\n\nThese terms and our Privacy Policy are the whole agreement between you and QDev about the products. If part of them can't be enforced, the rest still applies. If we don't enforce something straight away, we haven't given up the right to. You can't transfer these terms to someone else; we can, as part of a merger, sale or reorganization.\n\n## Contact\n\nQuestions about these terms: [info@quantumdev.io](mailto:info@quantumdev.io).\nHelp with an order or account: [support@quantumdev.io](mailto:support@quantumdev.io).\n\nQDev Inc., a Delaware corporation.\n"
  },
  {
   "id": "page:/changelog/",
   "type": "page",
   "title": "Changelog",
   "url": "https://quantumdev.io/changelog/",
   "markdownUrl": "https://quantumdev.io/changelog/index.md",
   "description": "What changed on quantumdev.io, newest first.",
   "updated": "2026-11-21",
   "text": "# Changelog\n\nCanonical: https://quantumdev.io/changelog/\n\nWhat changed on quantumdev.io, newest first.\n\n## Organisers can claim their event\n\n2026-11-21, https://quantumdev.io/changelog/#2026-11-21-claim-your-event\n\nEvery listed event's page now asks \"Run this event? Claim this listing.\"\n\n- **Prove it with a code.** Give an address at the site the listing links to. We email a code there, you type it in, and the listing is yours. We do not keep the address.\n- **Or tell us.** Where the listing links to a site many organisers share, such as a ticketing page, a code proves nothing. Tell QDev why the listing is yours and we decide.\n- **See your numbers.** \"Events you organise\" in My events shows how many members keep your event and how many are going. Two numbers, never names.\n- **Corrections that carry weight.** A correction you send to your own event is marked as the organiser's. QDev still checks each one.\n- **Named on the page.** If your profile is public, the event's page names you as its organiser. Give the claim up at any time.\n\n## Add slides and recordings to a past event\n\n2026-11-21, https://quantumdev.io/changelog/#2026-11-21-links-from-an-event\n\nFrom the day an event begins, its page has a new section, \"From the event\".\n\n- **Add a link.** \"Add a link\" opens a short form in My events: the address, what it is (slides, a recording, a write-up or something else) and a title for it.\n- **Checked first.** QDev opens each link, with help from an AI agent, before it shows. An approved link shows on the event's page without your name.\n- **Spoke at.** Tick \"This is my own talk\" and, where your profile shows your events, it lists the talk under \"Spoke at\".\n- **Yours to take down.** \"What you sent\" lists each link with its status, and \"Take it down\" removes one that is showing.\n\n## A past event keeps its page\n\n2026-11-21, https://quantumdev.io/changelog/#2026-11-21-past-events-keep-their-page\n\nUntil now an event's page answered \"No event here\" from the day after it ended. It now stays for two years.\n\n- **A record of the event.** The page keeps the event's name, dates, place, summary and the organiser's link, and says the event is over.\n- **Nothing left to act on.** Save, I'm going, the deadlines and \"Suggest a correction\" are gone from a past event's page.\n- **Who went.** Members who chose to be listed on the event's page stay on it, under \"Who went\". Untick the box in My events to take yourself off.\n- **Links that keep working.** An event you went to links to its page from My events and from your profile, and search engines keep finding it.\n\n## Send an event, or a correction\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-send-an-event\n\nMy events has a form for sending QDev an event, and every event's page has a link to suggest a correction.\n\n- **Send an event.** Give its name, the organiser's page and its first day, and whatever else you know. QDev reads it, with help from an AI agent that checks events, before it goes on the list.\n- **Suggest a correction.** \"Suggest a correction\" on an event's page opens the same form on that event. Change what is wrong, or write a note.\n- **See what happened.** \"What you sent\" lists each one as waiting, approved, not listed (with QDev's reason) or withdrawn. You can withdraw one while it waits.\n- **Credit.** When we list an event you sent, its page says a member added it, and names you if your profile is public.\n\n## Show your events, and see who's going\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-events-on-your-profile\n\nMy events has a new section, What others see, with two boxes. Both are off until you tick them.\n\n- **Events on your profile.** Tick the first and your public profile gains an Events section: the events you marked \"I'm going\", and later the ones you went to. Events you only saved are never shown.\n- **Who's going.** Tick the second and your name, what you are open to and a link to your profile go on the page of each event you are going to. A list appears once three members have joined it.\n- **Yours to undo.** Untick either box and your events or your name come down within a few minutes. You are listed only while your profile is public.\n\nThe Privacy Policy says what each box shows and what we keep to make the lists.\n\n## Hear when a new event fits you\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-new-events-that-fit\n\nMy events has a second email you can ask for.\n\n- **New events that fit you.** Tick \"Email me when a new event fits my profile\" and we email you when an event new to the list is in your city or country, or names an area, tool or kind of hardware on your profile.\n- **Not often.** At most one email a day, and only when something new fits. An event tagged loosely like your interests is not enough to email you.\n- **Asked for apart.** This box and the deadline reminders are separate, both off until you tick them. The link in any email turns both off.\n\nThe Privacy Policy says what we compare and what we keep.\n\n## Events picked for you\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-picked-for-you\n\nMy events has a new list, Picked for you.\n\n- **What it picks.** Upcoming events in your city or country, and events about the areas, tools and hardware on your profile. Each says why it was picked, and one press saves it.\n- **What it leaves out.** Events you already keep, and events that fit nothing on your profile. The list is never filled for the sake of it.\n- **Where it happens.** In your own browser, from your profile and the public events list. Nothing about you is sent anywhere to make a pick.\n\nIf the list is empty, add your city or the areas you work in to your profile.\n\n## Deadline reminders by email\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-deadline-reminders\n\nMy events can now email you before a deadline passes.\n\n- **Off until you ask.** Tick \"Email me before each deadline of the events I keep\" in My events. Until you do, we send nothing.\n- **What you get.** One email about a week before and one the day before each dated deadline of an event you saved or are going to, with the event, what closes or is due, and where we read it.\n- **Where it goes.** To the address you sign in with, and nowhere else.\n- **Stopping.** Untick the box, or use the link in any reminder. The link asks first and needs no sign-in.\n\nThe Privacy Policy says what we keep to do this.\n\n## A page for every event\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-a-page-for-every-event\n\nEach event on the Events page now has an address of its own, such as [quantumdev.io/e/aws-reinvent-2026](https://quantumdev.io/e/aws-reinvent-2026).\n\n- **Everything about one event on one page.** Its dates, its place, the organiser's link, and each deadline with the page we read it from and the day we checked.\n- **Made to be shared.** A link to an event's page previews with the event's name, days and city, and Copy link on the Events page now copies that address.\n- **Save it there.** Signed in, the page has the same Save and I'm going as the Events page, and Add to calendar works for everyone.\n\nEach event is run by its own organiser, and its page says so. When an event is over, its page comes down with it.\n\n## Every quantum talk at AWS re:Invent\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-quantum-talks-at-reinvent\n\nAWS re:Invent 2026 is on the Events page, and its quantum sessions have a page of their own.\n\n- **Every quantum session in one place.** The [talks page](https://quantumdev.io/events/talks/aws-reinvent-2026/) lists the 18 sessions about quantum in AWS's catalog: 7 on quantum computing, from Amazon Braket hardware to error correction and chemistry, and 11 on post-quantum cryptography.\n- **What each one has.** Its day, time and room in Las Vegas, AWS's own description, its speakers with a link to their LinkedIn where we could confirm it, and a link to the session in AWS's catalog.\n- **On the Events page.** re:Invent has a line under the clocks until its last day, and its entry links to the talks.\n\nre:Invent is run by Amazon Web Services, not by QDev. The descriptions are AWS's words, copied on 20 November. Times and rooms change, so check AWS's catalog before you go. quantumdev.io runs on AWS.\n\n## Save events and keep their deadlines\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-my-events\n\nThe Events page and your account now work together.\n\n- **Save an event, or say you are going.** Signed in, every event on the Events page has two buttons, Save and I'm going. Kept events are marked in the list.\n- **My events.** Your account has a new Events tab with what you are going to and what you saved, soonest first, each with its next deadline, such as an early rate closing or a call for papers.\n- **Your own calendar address.** Ask for one in My events and subscribe to it in your calendar app. It holds every event you kept and each of their deadlines as a day of its own, and it updates as you save events and as their dates change. Anyone who has the address can read your events, so you can replace it or turn it off at any time.\n\nOnly you can see the events you keep. The Privacy Policy says what is stored.\n\n## Confirm your password at sign-up\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-confirm-your-password\n\nCreating an account now asks for your password twice, so a slip of the finger cannot lock you out of a new account.\n\n- A sign under the password turns from red to green once it is long enough.\n- A second sign under the confirm field turns green once the two passwords are the same.\n\n## Profile edits are kept\n\n2026-11-20, https://quantumdev.io/changelog/#2026-11-20-profile-edits-are-kept\n\nSaving one section of your profile no longer discards what you typed in another, and two saves made close together both go through.\n\nIf your profile fails to load, the editor now says so and stays locked, so nothing can be saved over it.\n\n## A public changelog\n\n2026-11-19, https://quantumdev.io/changelog/#2026-11-19-a-public-changelog\n\nEvery change to quantumdev.io is listed on this page, newest first, and the [RSS feed](https://quantumdev.io/changelog/feed.xml) has every one. New changes are also posted on [QDev's X account](https://x.com/quantumdevio), which every page now links to from its footer.\n\n## Public profiles\n\n2026-11-19, https://quantumdev.io/changelog/#2026-11-19-public-profiles\n\nYour profile can now be public, at quantumdev.io/u/ followed by the handle you choose.\n\n- It stays private until you switch it to public.\n- You choose which parts a visitor sees, and whether search engines may list it and AI assistants may read it.\n- It shows your member number, in the order members joined.\n- Every public profile has a link to report it.\n\nA public profile is never offered for training an AI model.\n\n## Account codes from QDev\n\n2026-11-19, https://quantumdev.io/changelog/#2026-11-19-codes-from-qdev\n\nThe codes that confirm a new account or reset a password now come from QDev's own address, `account@updates.quantumdev.io`.\n\n## Profiles of your own\n\n2026-11-19, https://quantumdev.io/changelog/#2026-11-19-profiles\n\nEvery QDev has a profile.\n\n- A banner drawn for you alone, in one of seven quantum themes, and a default picture of your own qubit. You can also upload a picture and a banner.\n- A headline, what you are working on now, what you are open to, and links to your pages elsewhere.\n- Your tools, the hardware you have run on, and what you work on.\n- Your experience and education.\n\nYou can download everything your profile holds from your account's settings.\n\n## Become a QDev\n\n2026-11-19, https://quantumdev.io/changelog/#2026-11-19-become-a-qdev\n\nYou can now make a free QDev account with an email address and a password, and an authenticator app if you want one. \"Become a QDev\" is in the menu of every page, and the email list has moved to the footer.\n\n## Event deadlines\n\n2026-11-18, https://quantumdev.io/changelog/#2026-11-18-event-deadlines\n\nEvents on the [Events](https://quantumdev.io/events/) page show their deadlines, such as calls for papers and registration, where the organiser publishes them.\n\n## Listed in the official MCP registry\n\n2026-11-18, https://quantumdev.io/changelog/#2026-11-18-listed-in-the-mcp-registry\n\nQDev's MCP server is listed in the official MCP registry as `io.quantumdev/site`, so AI tools that browse the registry can find it.\n\n## An events calendar feed\n\n2026-11-15, https://quantumdev.io/changelog/#2026-11-15-an-events-calendar-feed\n\nYou can subscribe to the events in your own calendar app. [quantumdev.io/events/events.ics](https://quantumdev.io/events/events.ics) updates whenever the catalog does.\n\n## Open to AI assistants\n\n2026-11-14, https://quantumdev.io/changelog/#2026-11-14-open-to-ai-assistants\n\nAI assistants can read quantumdev.io directly.\n\n- Every page has a markdown copy, and `site.json` holds them all.\n- The published events come with a JSON Schema.\n- An MCP server at `quantumdev.io/mcp` answers questions about the events, pages and posts, and can start an email list sign-up.\n\nTraining an AI model on the site needs a licence, priced in the [Terms](https://quantumdev.io/terms/#ai).\n\n## The blog\n\n2026-11-14, https://quantumdev.io/changelog/#2026-11-14-the-blog\n\nThe [QDev Blog](https://quantumdev.io/blog/) is open, with writing on quantum computing. Posts can carry mathematics, code and diagrams, and an [RSS feed](https://quantumdev.io/blog/feed.xml) has every one.\n\n## Find events by place and date\n\n2026-11-10, https://quantumdev.io/changelog/#2026-11-10-find-events-by-place\n\nThe [Events](https://quantumdev.io/events/) page finds quantum events by continent, country, state and city on one screen.\n\n- A calendar of the months ahead.\n- The events nearest a region you choose.\n- For an event you pick, the others that fit the same trip.\n\nThe events are run by their own organisers; QDev keeps the catalog.\n\n## The home page tells one story\n\n2026-11-10, https://quantumdev.io/changelog/#2026-11-10-the-home-page-tells-a-story\n\nThe home page now moves as you scroll. The four products come first, with a quantum circuit drawn across them gate by gate, then what QDev is, the next event and the email list.\n\n## Events on a globe\n\n2026-11-06, https://quantumdev.io/changelog/#2026-11-06-events-on-a-globe\n\nThe [Events](https://quantumdev.io/events/) page opens on a globe of quantum computing events around the world.\n\n## The email list\n\n2026-11-04, https://quantumdev.io/changelog/#2026-11-04-the-email-list\n\nYou can join QDev's email list from any page. A link in your inbox confirms it, and from then on you get one email when each product opens and when an event is announced. Every one has an unsubscribe link.\n\n## Privacy Policy and Terms\n\n2026-11-01, https://quantumdev.io/changelog/#2026-11-01-privacy-policy-and-terms\n\nThe [Privacy Policy](https://quantumdev.io/privacy/) and the [Terms](https://quantumdev.io/terms/) are published, and every page links to them from its footer.\n\n## quantumdev.io, rebuilt\n\n2026-10-31, https://quantumdev.io/changelog/#2026-10-31-quantumdev-io-rebuilt\n\nQDev moved to quantumdev.io, rebuilt on the Quantum Design System, with one page for the four products in a light and a dark theme.\n\n- [Quantum Shop](https://shop.quantumdev.io) is open.\n- Quantum Learner, Quantum Architect and Quantum Instructor are marked coming soon.\n"
  },
  {
   "id": "page:/events/talks/aws-reinvent-2026/",
   "type": "page",
   "title": "Quantum talks at AWS re:Invent 2026",
   "url": "https://quantumdev.io/events/talks/aws-reinvent-2026/",
   "markdownUrl": "https://quantumdev.io/events/talks/aws-reinvent-2026/index.md",
   "description": "Every session about quantum at AWS re:Invent 2026, 30 November to 4 December 2026 in Las Vegas: 7 on quantum computing and 11 on post-quantum cryptography, each with Amazon Web Services' description, its speakers, and its time and room.",
   "updated": "2026-11-20",
   "text": "# Quantum talks at AWS re:Invent 2026\n\nCanonical: https://quantumdev.io/events/talks/aws-reinvent-2026/\n\nEvery session about quantum at AWS re:Invent 2026, 30 November to 4 December 2026 in Las Vegas: 7 on quantum computing and 11 on post-quantum cryptography, each with Amazon Web Services' description, its speakers, and its time and room.\n\nAWS re:Invent 2026 is run by Amazon Web Services, not by QDev. The titles and the descriptions are Amazon Web Services' own words, copied on 2026-11-20 from https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=quantum. Times are local to Las Vegas (America/Los_Angeles).\n\n## Quantum computing\n\nSessions AWS files under Quantum Technologies: Amazon Braket, hardware, error correction, chemistry and quantum software.\n\n### CMP340: Drug discovery and materials use cases with Amazon Braket and Q-CTRL\n\nChalk talk, 300 – Advanced\n\n- Monday 30 November, 13:00 to 14:00, MGM Grand | Level 1 | Boulevard 163\n\n> Classically hard problems such as simulating strongly correlated electrons and 1D NMR spectra, critical to material and drug discovery, can be sped up on quantum computers but until now had not been demonstrated. Q-CTRL shows how Fire Opal on Amazon Braket enables simulating 1D NMR spectra for spin systems up to 34 qubits and achieved a 3,000x speedup on the Fermi-Hubbard model at 120 qubits, demonstrating practical quantum advantage. This chalk talk unpacks how AWS hybrid workflows enabled these outcomes and what enterprises should do now to capture ROI with quantum.\n\n- [Nick White](https://www.linkedin.com/in/nicholas-white-74b18a120/), Partner Solutions Architect - Quantum Partner Strategy Lead, Amazon Web Services\n- [Alex Shih](https://www.linkedin.com/in/alexshih/), VP of Product, Q-CTRL\n\n[CMP340 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=CMP340)\n\n### INV518: Quantum Error Correction - Progress, Hardware, and What's Next\n\nChalk talk, 500 – Distinguished\n\n- Wednesday 2 December, 08:30 to 09:30, MGM Grand | Level 3 | Premier 311\n\n> Quantum computers have the potential to solve some of today's most challenging computational problems — but they suffer from significantly higher error rates than their classical counterparts. In this talk, we introduce quantum error correction (QEC) and compare the hardware requirements of leading QEC methods against the commercial roadmaps of quantum hardware companies. To ground the discussion, we begin with a brief revisit of classical error correction before walking through recent breakthroughs in quantum error correction. Attendees will leave with an informed perspective on where QEC stands today and the challenges that will define the field's progress in the coming years.\n\n- [Tyler Takeshita](https://www.linkedin.com/in/tyler-takeshita/), Sr. Applied Scientist, Amazon Web Services\n- [Mao Lin](https://www.linkedin.com/in/maolin2/), Applied Scientist, Amazon Web Services\n\n[INV518 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=INV518)\n\n### CMP349: Your next Quantum developer might be an AI agent\n\nChalk talk, 300 – Advanced\n\n- Wednesday 2 December, 12:00 to 13:00, MGM Grand | Level 1 | Boulevard 165\n\n> AI is changing who can build quantum applications. See how AI assistants including Kiro, Claude Code, Codex, Cursor, Jupyter AI, and Classiq help developers move from natural-language prompts to optimized quantum workloads on Amazon Braket. Learn how AI generates code, synthesizes hardware-aware circuits, benchmarks on simulators and QPUs, and integrates with HPC resources. Through real examples, you'll see how AI-native workflows reduce complexity, accelerate experimentation, and make quantum development accessible to more builders.\n\n- [Nick White](https://www.linkedin.com/in/nicholas-white-74b18a120/), Partner Solutions Architect - Quantum Partner Strategy Lead, Amazon Web Services\n- [Ishaan Pakrasi](https://www.linkedin.com/in/ishaanpakrasi/), Senior Product Manager - Technical, Amazon Web Services\n- Erik Garcell, Director of Quantum Solutions Architecture - Americas, Classiq\n\n[CMP349 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=CMP349)\n\n### CMP402: Build hybrid quantum-classical solutions for science and engineering on Amazon Braket\n\nWorkshop, 400 – Expert\n\n- Thursday 3 December, 08:30 to 10:30, MGM Grand | Level 1 | Grand 116\n\n> Take the next step into hybrid quantum computing. This expert-level workshop combines Amazon Braket and AWS Parallel Computing Service to build hybrid quantum-classical resources that enable scientific and engineering research in fields like physics and chemistry. You learn how to structure hybrid programs, manage classical resources, and interpret results from real quantum devices. Expect hands-on exercises, detailed guidance, and examples you can use as a foundation for your own projects.\n\n- [Tyler Takeshita](https://www.linkedin.com/in/tyler-takeshita/), Sr. Applied Scientist, Amazon Web Services\n- [Sebastian Stern](https://www.linkedin.com/in/sebastianstern/), Principal Specialist Solutions Architect Quantum, Amazon Web Services\n\n[CMP402 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=CMP402)\n\n### CMP310: Beyond the simulator: Quantum chemistry workflows on real hardware\n\nBuilders' session, 300 – Advanced\n\n- Thursday 3 December, 11:30 to 12:30, MGM Grand | Level 3 | Chairman's 363 | Content Hub | Builders' Session 2\n\n> Get hands-on experience running quantum computing experiments using the Kvantify SDK; the same workflow used by researchers. In this workshop, participants will set up and execute computational workflows, testing them locally, on a classical simulator via Amazon Braket, and finally on a real quantum processing unit (QPU). Through practical exercises in enzyme engineering and battery development, learners will see how quantum computing accelerates discovery in life sciences and materials science, bridging the gap from simulation to real quantum hardware.\n\n- [Tyler Takeshita](https://www.linkedin.com/in/tyler-takeshita/), Sr. Applied Scientist, Amazon Web Services\n- [Nick White](https://www.linkedin.com/in/nicholas-white-74b18a120/), Partner Solutions Architect - Quantum Partner Strategy Lead, Amazon Web Services\n- [Veronika Liutarevich](https://www.linkedin.com/in/veronika-liutarevich-137977254/), Solutions Architect, Amazon Web Services\n- [Sebastian Stern](https://www.linkedin.com/in/sebastianstern/), Principal Specialist Solutions Architect Quantum, Amazon Web Services\n- [Nicola Smyth](https://www.linkedin.com/in/nicola-smyth-5b5a44103/), Startup Solutions Architect, Amazon Web Services\n\n[CMP310 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=CMP310)\n\n### CMP348: Building better quantum computers with digital twins\n\nChalk talk, 300 – Advanced\n\n- Thursday 3 December, 11:30 to 12:30, MGM Grand | Level 1 | Boulevard 163 (CMP348-R)\n- Friday 4 December, 11:30 to 12:30, Wynn/Encore | Convention Promenade | Montrachet 1 (CMP348-R1)\n\n> Quantum computers could revolutionize fields like pharmacy and logistics, but today's hardware is too error-prone for practical use. To design better error-correction strategies, simulating how real quantum hardware operates is essential, yet computationally intractable. In this talk, we demonstrate quantum processor digital twins developed by Quantum Elements and powered by AWS HPC infrastructure - that faithfully mimic real hardware at experimental scale. Attendees will leave understanding why error correction matters and how classical compute accelerates the path to useful quantum computing.\n\n- [Tong Shen](https://www.linkedin.com/in/tong-shen-872a25170/), Research Scientist, Quantum Elements\n- [Benchen Huang](https://www.linkedin.com/in/benchen-huang/), Specialist Solutions Architect Quantum Tech, Amazon Web Services\n\n[CMP348 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=CMP348)\n\n### CMP306: Delivering next-generation quantum hardware on Amazon Braket\n\nBreakout session, 300 – Advanced\n\n- Thursday 3 December, 12:00 to 13:00, Caesars Palace | Promenade Level | Roman I\n\n> In this session, we discuss bringing next-generation quantum devices to Amazon Braket and dive deep on how our hardware and software capabilities unlock fault-tolerant quantum computing for our customers. We share early results and outline how organizations can begin preparing today for the fault-tolerant era.\n\n- [Eric Kessler](https://www.linkedin.com/in/eric-kessler-aws/), General Manager, Advanced and Scientific Compute, Amazon Web Services\n\n[CMP306 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=CMP306)\n\n## Post-quantum cryptography\n\nSessions on moving to encryption that a quantum computer cannot break.\n\n### COM210: Post-Quantum Readiness on AWS: Where Do You Start?\n\nMeetup, 200 – Intermediate\n\n- Monday 30 November, 11:00 to 11:45, Wynn/Encore | Upper Convention Promenade | Cristal 2 | Content Hub | AWS Meetups\n\n> Data encrypted today could be decrypted later by a future quantum computer, and the move to post-quantum cryptography is already underway across AWS. This practical meetup brings together security and platform engineers to compare how they are preparing. We'll discuss building a cryptographic inventory, where AWS already supports hybrid post-quantum TLS on service endpoints such as AWS KMS, AWS Certificate Manager, and AWS Secrets Manager, and what that means for your clients and SDKs. We'll also cover designing for crypto agility so algorithms can change without rewrites, and deciding which data and systems to address first. Come at any stage, share what you've tried, and leave with a practical starting plan.\n\nNo speaker is named yet.\n\n[COM210 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=COM210)\n\n### GHJ203: Post-quantum cryptography tabletop experience\n\nGamified learning, 200 – Intermediate\n\n- Tuesday 1 December, 12:00 to 14:00, Wynn/Encore | Upper Convention Promenade | Bollinger (GHJ203-R)\n- Thursday 3 December, 15:00 to 17:00, Wynn/Encore | Upper Convention Promenade | Bollinger (GHJ203-R1)\n\n> Quantum computing is advancing—is your cryptographic infrastructure ready? In this immersive tabletop, executives will confront the urgent reality that current public-key cryptographic standards face obsolescence. Make high-stakes decisions on migration timelines, technology selection, and stakeholder coordination as quantum threats loom over customer data, transaction systems, and competitive intelligence. Balance budget constraints against discovery, risk prioritization, and implementation speed while navigating regulatory requirements and operational continuity. Compete to build quantum-resistant infrastructure before adversaries exploit today's vulnerabilities. Leave with actionable strategies for leading a post-quantum cryptography migration that protects trust and secures competitive advantage. Walk-ins welcome.\n\nNo speaker is named yet.\n\n[GHJ203 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=GHJ203)\n\n### INV529: Beyond RSA: Post-quantum data encryption using ML-KEM in AWS KMS\n\nChalk talk, 500 – Distinguished\n\n- Tuesday 1 December, 12:00 to 13:00, MGM Grand | Level 3 | Room 353\n\n> RSA and Elliptic Curve Cryptography secure nearly every application today, and both are vulnerable to future cryptographically relevant quantum computers. Data captured now can be decrypted later, so applications needing long-term confidentiality should adopt post-quantum encryption well before those computers arrive. AWS KMS now offers ML-KEM, the NIST-standardized replacement for RSA and ECDH key establishment and ultimately encryption under asymmetric keys. It protects the private keys in FIPS 140-3 Level 3 HSMs. In this session, learn about ML-KEM and how to use it as a building block for quantum-safe encryption using Hybrid Public Key Encryption (HPKE).\n\n- Vipul Gupta, Principal Engineer (AWS KMS), Amazon Web Services\n\n[INV529 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=INV529)\n\n### SEC201: Turn PQC readiness into the trust story your board needs\n\nChalk talk, 200 – Intermediate\n\n- Wednesday 2 December, 08:30 to 09:30, Wynn/Encore | Convention Promenade | Montrachet 1 (SEC201-R)\n- Wednesday 2 December, 16:00 to 17:00, Wynn/Encore | Level 1 | Chopin 2 (SEC201-R1)\n\n> Organizations planning post-quantum cryptography (PQC) migrations face the same three questions: where do I start my migration plan, which workloads do I prioritize, and how do I get buy-in from leadership? With NIST PQC standards finalized, the algorithm question is settled, the open question is how to plan and fund the transition. In this chalk talk we whiteboard the PQC Readiness Maturity Model: five levels from discovery to quantum-ready, each with measurable milestones. You leave with a practical framework to scope your inventory, prioritize transitions, modernize how you protect data, and build the board narrative that funds your migration program.\n\n- Angela Morrison, HCLS Professional Services Practice Leader, Amazon Web Services\n- [Jonathan Jenkyn](https://www.linkedin.com/in/jjenkyn/), Snr Security and Compliance SA, AWS Security Assurance Services\n\n[SEC201 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=SEC201)\n\n### INV524: Verifying Post-Quantum Cryptography at Speed\n\nChalk talk, 500 – Distinguished\n\n- Wednesday 2 December, 10:00 to 11:00, Wynn/Encore | Convention Promenade | Latour 5\n\n> Traditionally, cryptographic engineering involves a performance/assurance trade-off: clean code is easy to audit and maintain, yet slow; handwritten assembly is fast, yet hard to validate. Discover how automated reasoning helps to overcome this tension, following the example of mlkem-native and mldsa-native, two open-source post-quantum cryptography libraries co-maintained by AWS and deployed in AWS-LC, Amazon's cryptography library. In these libraries, constraint solving enables automated superoptimization at the assembly level, while formal verification vets the result. Contrary to the popular belief that assurance costs speed, it is automated reasoning that enables deep optimization, verification and rapid deployment of cryptographic code at cloud scale.\n\n- [Hanno Becker](https://www.linkedin.com/in/hanno-becker-a41106101/), Principal Applied Scientist, Amazon Web Services\n\n[INV524 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=INV524)\n\n### SEC293: Data protection and the post-quantum shift\n\nMeetup, 200 – Intermediate\n\n- Wednesday 2 December, 12:00 to 12:45, Wynn/Encore | Upper Convention Promenade | Cristal 2 | Content Hub | AWS Meetups\n\n> Encryption, key and secrets management, and data sovereignty all face new pressure as post-quantum timelines compress. Bring your plans for post-quantum migration, secrets rotation at scale, and protecting sensitive data in regulated and sovereign environments. Compare timelines and approaches with others who are planning the same moves.\n\nNo speaker is named yet.\n\n[SEC293 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=SEC293)\n\n### SEC345: Future-proof your keys: HSMs to AWS KMS in the post-quantum era\n\nChalk talk, 300 – Advanced\n\n- Thursday 3 December, 08:30 to 09:30, Wynn/Encore | Lower Convention Promenade | Meursault 2\n\n> Cryptography is changing fast. The dedicated HSM you rely on today, whether AWS CloudHSM or an on-premises appliance, carries operational overhead that will only grow as the industry moves toward post-quantum cryptography (PQC) and stricter digital sovereignty requirements. AWS Key Management Service (KMS) now covers most workloads that once required dedicated HSMs, with the same FIPS 140-3 Level 3 security, lower cost, and zero infrastructure management, while giving you a clear on-ramp to PQC and the sovereignty controls to keep your keys and data under your control. In this chalk talk, bring your architecture and we'll work through it together: we'll assess your workload, map signature and encryption migration paths, and show how moving to KMS puts you on the fastest path to post-quantum readiness. Meet the AWS Cryptography team and leave with a plan and concrete next steps.\n\n- Kevin Kevin, Sr. Product Manager - Technical, Amazon Web Services\n- [Derek Tumulak](https://www.linkedin.com/in/tumulak/), Principal Product Manager, Amazon Web Services\n\n[SEC345 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=SEC345)\n\n### SEC205: Journey to post-quantum (PQ) authentication in AWS\n\nChalk talk, 200 – Intermediate\n\n- Thursday 3 December, 12:00 to 13:00, Wynn/Encore | Lower Convention Promenade | Fleurie\n\n> You use certificates for identity, keys for token signing, and long-lived roots of trust. Now you need to migrate to post-quantum cryptography (PQC) for authentication. In this session, learn how AWS helps you start and stay PQ-ready while preserving backwards compatibility, performance, and AI-friendly security by default. We will demystify the present state and future trajectory of PQ authentication. We discuss the updated CNSA 2.0 dates when RSA and EC will no longer satisfy authentication requirements and the uncertainty around PQ public certificate formats and Merkle Trees in non-browser scenarios.\n\n- [Avni Rambhia](https://www.linkedin.com/in/avnirambhia/), Principal Product Manager, Amazon Web Services\n- [Oren Nachman](https://www.linkedin.com/in/oren-nachman/), Senior Manager, AWS Certificates & PKI (ACM, PCA, Signer), Amazon Web Services\n\n[SEC205 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=SEC205)\n\n### SEC357: Quantum-proof your IoT fleet: migrate devices to Post-quantum cryptography OTA\n\nWorkshop, 300 – Advanced\n\n- Thursday 3 December, 12:30 to 14:30, Wynn/Encore | Upper Convention Promenade | Cristal 3\n\n> NIST has announced post-quantum cryptography (PQC) standards and CNSA 2.0 requires quantum-resistant algorithms by 2027. IoT devices deployed today operate for 10–20 years, so they need to be quantum-resistant from day one. In this workshop, migrate a real ESP32-S3 SoC from classical TLS to post-quantum cryptography over the air with zero physical access required. Connect to AWS IoT Core, trigger an OTA delivering ML-KEM and ML-DSA with quantum-safe certificates from AWS Private CA, and watch your device reboot into PQC-secured MQTT. No embedded experience needed. You leave with the ESP32-S3 and open-source code to quantum-proof your fleet at scale.\n\n- [Pravin Nair](https://www.linkedin.com/in/awsnair/), Sr. Security Strategy SA, Amazon Web Services\n- Samuel Waymouth, Senior Compliance Specialist, Amazon Web Services\n\n[SEC357 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=SEC357)\n\n### SEC316: Crypto escape room: race to migrate to quantum resistance\n\nBuilders' session, 300 – Advanced\n\n- Thursday 3 December, 16:00 to 17:00, Wynn/Encore | Upper Convention Promenade | Cristal 2 | Content Hub | Builders' Session 1\n\n> The migration to post-quantum cryptography is underway and here's a chance to hone your skills. In this gamified builders' session, each team races a countdown clock through different cryptographic migration challenges: TLS endpoint migration, ML-DSA key for code signing, quantum ready PKI, and an audit of your PQ posture. Teams complete quests using the AWS Key Management Service (KMS), AWS Private Certificate Authority, and AWS Secrets Manager services. A live leaderboard tracks progress across teams. Leave with new skills and the confidence to build quantum-resistant cryptographic infrastructure.\n\n- [Jay Viera](https://www.linkedin.com/in/vierajay/), Solutions Architect, Amazon Web Services\n- [Phoebe Sorkin](https://www.linkedin.com/in/phoebe-sorkin/), Solutions Architect, Amazon Web Services\n- [Arun Krishnaswamy](https://www.linkedin.com/in/krishnaswamyarun/), Principal Solutions Architect, Amazon Web Services\n- Shankar Rajagopalan, Sr. Solutions Architect, Amazon Web Services\n- [Margo Cronin](https://www.linkedin.com/in/margo-cronin-478407/), Principal Solutions Architect, Amazon Web Services\n\n[SEC316 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=SEC316)\n\n### SEC204: Post-quantum readiness at scale with AWS and DXC Technology\n\nBreakout session, 200 – Intermediate\n\n- Friday 4 December, 11:30 to 12:30, Wynn/Encore | Upper Convention Promenade | Cristal 2 | Content Hub | Blue Theater\n\n> Post-quantum migrations succeed when organizations can discover cryptographic assets, assign ownership, and continuously manage risk. Organisations are facing a rapid increase in machine identities, service identities, APIs, containers, AI workloads, and other cryptographic assets. At the same time, certificate lifetimes continue to shrink, making manual lifecycle management increasingly unsustainable. AWS and DXC Technology show how enterprise governance connects with crypto agility. See how DXC built a cryptographic asset intelligence and orchestration platform on AWS to discover and correlate assets across AWS and enterprise PKI, automate certificate lifecycles, and assess post-quantum readiness. A live proof of concept demonstrates inventory, prioritization, and continuous reassessment. Leave with an operating model aligned with NIST and CNSA 2.0 timelines.\n\n- Rushir Patel, Head of Data Protection GTM, Amazon Web Services\n- [Douglas Skirving](https://www.linkedin.com/in/douglasskirving/), Global Lead Key & Certificate Services - GTCO, DXC Technology\n\n[SEC204 in AWS's session catalog](https://registration.awsevents.com/flow/awsevents/reinvent2026/eventcatalog/page/eventcatalog?search=SEC204)\n"
  },
  {
   "id": "page:/blog/",
   "type": "page",
   "title": "QDev Blog",
   "url": "https://quantumdev.io/blog/",
   "markdownUrl": null,
   "description": "Writing on quantum computing from QDev Inc., the company behind The Quantum Environment.",
   "updated": "2026-11-04",
   "text": "# QDev Blog\n\nCanonical: https://quantumdev.io/blog/\n\nWriting on quantum computing from QDev Inc., the company behind The Quantum Environment.\n\n- [When AI Writes Quantum Code, Who Checks the Work?](https://quantumdev.io/blog/when-ai-writes-quantum-code-who-checks-the-work/), 2026-11-04: What an ORNL audit of quantum code benchmarks teaches builders about AI-generated circuits.\n- [Teaching AI to Catch Quantum Errors](https://quantumdev.io/blog/teaching-ai-to-catch-quantum-errors/), 2026-10-30: How AWS trained NVIDIA's AI pre-decoder for quantum error correction on EC2 GPUs, cutting decoding latency 2.4 to 2.9 times, and what it means for builders.\n- [Quantum Advantage](https://quantumdev.io/blog/quantum-advantage/), 2026-09-06: What quantum advantage means, which problems to try first, and how AWS builders can start this week, with a fault-tolerant QuEra system coming to Amazon Braket.\n- [Navigating Quantum Innovation](https://quantumdev.io/blog/navigating-quantum-innovation/), 2026-08-19: Michael Brett of Amazon Braket on what is easy and hard in quantum computing, and where AWS builders should focus: algorithms, benchmarks and AI.\n- [Entanglement on Amazon Braket: The CNOT Gate and Your First Bell State](https://quantumdev.io/blog/entanglement-on-amazon-braket-the-cnot-gate-and-your-first-bell-state/), 2026-07-19: A hands-on guide to quantum entanglement on Amazon Braket. Build a Bell state with the CNOT gate, run it in 2 lines, and measure the correlation.\n- [A Developer's Field Guide to Amazon Braket](https://quantumdev.io/blog/a-developers-field-guide-to-amazon-braket/), 2026-07-17: A guided tour of the Amazon Braket console, page by page. What each button does, when you'd use it, and how to run your first task on a real quantum computer.\n- [Unlocking Superposition: The Hadamard Gate on Amazon Braket](https://quantumdev.io/blog/unlocking-superposition-the-hadamard-gate-on-amazon-braket/), 2026-06-28: Learn to build your first quantum circuit on AWS. Master the Hadamard gate, explore superposition, and get started with Amazon Braket.\n- [Fault Tolerance: Quantum's Hardest Problem Has an End-Date](https://quantumdev.io/blog/fault-tolerance-quantums-hardest-problem-has-an-end-date/), 2026-06-20: Quantum's real bottleneck isn't qubit count. It's fault tolerance. How error correction, AWS Ocelot, and QuEra's Libra are ending the noisy era by 2028.\n- [Dear Builder, Welcome to Quantum Computing](https://quantumdev.io/blog/dear-builder-welcome-to-quantum-computing/), 2026-06-17: An invitation to start exploring what is possible with quantum computing on Amazon Braket.\n- [The State of the Tech: Quantum Computing on AWS (2026)](https://quantumdev.io/blog/the-state-of-the-tech-quantum-computing-on-aws-2026/), 2026-06-01: Where quantum computing stands in 2026: what today's hardware can do, its limits, and how to build hybrid optimization and machine learning workflows on AWS.\n"
  },
  {
   "id": "post:when-ai-writes-quantum-code-who-checks-the-work",
   "type": "post",
   "title": "When AI Writes Quantum Code, Who Checks the Work?",
   "url": "https://quantumdev.io/blog/when-ai-writes-quantum-code-who-checks-the-work/",
   "markdownUrl": "https://quantumdev.io/blog/when-ai-writes-quantum-code-who-checks-the-work.md",
   "description": "What an ORNL audit of quantum code benchmarks teaches builders about AI-generated circuits.",
   "published": "2026-11-04",
   "updated": "2026-11-04",
   "author": "Christian Perez",
   "text": "# When AI Writes Quantum Code, Who Checks the Work?\n\nBy Christian Perez, published 2026-11-04\n\nCanonical: https://quantumdev.io/blog/when-ai-writes-quantum-code-who-checks-the-work/\n\nWhat an ORNL audit of quantum code benchmarks teaches builders about AI-generated circuits.\n\n> _Disclaimer: While this article covers a paper focused on Qiskit, AWS builders should know that the Qiskit-Braket provider lets you run Qiskit circuits on Amazon Braket simulators and quantum hardware._\n>\n> _These findings apply to your work as well._\n>\n> __\n>\n> _The final section provides more detail._\n\nOn one set of quantum computing tasks, a frontier large language model scored 0%. Then the researchers fixed the grader, left every line of the model's code untouched, and the same output scored 94.1%.\n\nHere is what \"fixed\" means. The task was T1, 17 Quantum Approximate Optimization Algorithm (QAOA) problems in a new benchmark the researchers built, and the model was GPT-6 Astra. The grader was checking exact resource quantities that the prompts never asked for. The researchers removed only those checks, kept every correctness check, and rescored the saved outputs. A second topic, T6, had the same kind of problem, and the same fix moved Astra from 1.2% to 92.9%.\n\nThat result comes from a new preprint out of the Quantum Science Center at Oak Ridge National Laboratory, \"Benchmarking the Natural Quantum Leap: Auditing LLM Evaluation for Quantum Code Generation\" (Seefried et al., arXiv 2610.28678). The team ran 15 models across 5 established quantum code benchmarks with 530 tasks, and 10 models on QuantumBenchEval, a new 96-task benchmark of their own. Then they did something that a lot of leaderboards actually tend to skip. They audited the graders, their own included.\n\nWhat I hope you get from all of this is that if you're picking an AI coding assistant for Qiskit or Braket work, or you're about to paste a model-generated circuit into your Amazon Braket project, this paper is a useful guide to what a score does and does not actually tell you about the LLM and how it behaves in the quantum ecosystem.\n\n## Why is quantum code harder to grade?\n\nWhen you grade normal code, a unit test asks a simple question: Did the function return the right value? Quantum code makes that question a little bit more complicated. A correct circuit can be written many ways. Two circuits can look nothing alike and still be the same operation up to a global phase. Outputs are often measurement distributions, so \"right\" can mean \"statistically close enough.\"\n\nThe paper opens with two examples of the work involved: preparing a four-qubit W state, and decomposing a CNOT into iSWAP gates. Tasks like these actually require reasoning about circuit semantics, measurement statistics, and backend constraints on top of ordinary API usage.\n\nBenchmark authors grade this in different ways. Some run unit tests, some ask another LLM to judge the answer, and some write custom scoring functions. The ORNL team ran every sweep through one harness and scored each with its own native evaluator.\n\n![Table of six quantum code benchmarks with their task counts and how they grade: QiskitHumanEval Standard, 151 tasks, unit tests; QiskitHumanEval Hard, 151, unit tests on the same tasks without the supplied code skeleton; QuanBench-44, 44, unit tests; QuanBench-117, 117, unit tests; QCoder, 67, an LLM judge with no executable tests; QuantumBenchEval (new), 96, task assertions plus an LLM rubric for the VQE topic, T3.](https://quantumdev.io/blog/media/image-03f31244d56a5ef419c0b7bfc7fe522863f24608-1260x668-webp-960.webp)\n\nEach model generated five completions per task from a shared prompt template, with no re-ranking, voting, or post-processing (Table 1 in Section 3 of the paper). Decoding settings were not identical across model providers: most models ran at a temperature of 0.8, and Claude 5.1 and Muse-Glimmer-30B ran at 1.0.\n\n## The Standard suite is pretty much saturated\n\n![Bar chart of QiskitHumanEval pass@1 for five models, from Table 2 of the paper. Every model scores far higher on Standard than on Hard: GPT-6 Astra about 98% against about 63%, Muse-Glimmer-30B about 95% against about 28%, and Claude Fable 5.1 about 70% against about 56%.](https://quantumdev.io/blog/media/image-0a991174a105e2d8f31cbcfede6d1c91ac2cddd5-1284x798-webp-960.webp)\n\nOn QiskitHumanEval Standard, GPT-6 Astra hit 98.7% pass@1, and Gemini 3.1 Pro Preview reached 97.6%.\n\n> _pass@k means the model passed the benchmark assessment at \"k\" number of attempts, therefore pass@1 means the model passed on the first attempt._\n\nAcross five attempts, only one of the 151 Standard tasks went unsolved by every model. Standard is 151 short programming problems on core primitives such as state preparation, circuit construction, and measurement. The prompt supplies the import and a function skeleton. An answer counts as correct if and only if it passes the task's unit tests. The strongest models now solve nearly all of it (Astra reaches 100% pass@1 in 8 of 10 task categories), so the authors describe the suite as near saturation and say the remaining headroom lies in harder tasks such as the Hard variant and their new benchmark. In a task-by-task comparison with Holm correction, Astra's 1.1-point lead over Gemini 3.1 Pro has an adjusted p of 0.90, which means standard cannot tell the two apart, not that they are proven equal.\n\n> _Holm correction is a way to avoid fooling yourself when you run several statistical tests at once. Each test has a small chance of showing a \"significant\" difference by pure luck. Run five tests, and the odds that at least one comes up by luck go up._\n>\n> _Holm Fixes that by raising the bar for each test. You sort your results from strongest to weakest. The strongest has to clear the strictest cutoff, and each one after it gets a slightly easier cutoff. Once one test fails, you stop and call the rest not significant too._\n>\n> _In the paper, the authors compared the top two models on five different test suites. After this correction, none of the gaps held up. That's why the article says the benchmarks can't really tell those top models apart._\n\nFor a builder, a high Standard score is fair evidence that a model handles short scaffolded Qiskit tasks, but it is not a reason to pick one top model over another, and it does not tell you how a model will do on harder work without the skeleton.\n\nQiskitHumanEval Hard looks very different. The best pass@1 drops to 63.8%, and 31 of 151 tasks went unresolved by every model. All 15 models scored lower on Hard than on Standard. Hard uses the same task IDs but takes away the import and function skeleton that Standard supplies, so the authors say the drop reflects sensitivity to prompt format and scaffolding, not only difficulty, and they treat it as a benchmark-level constraint rather than a measured difficulty effect. Within Hard, Astra ranges from 20.0% on quantum information to 85.7% on state preparation and analysis, though the authors caution that small categories have wide intervals.\n\n## First place is not really a clear win\n\nOn each of the five primary suites, the authors compared first and second place task by task. After Holm correction for all the model pairs compared within each suite, none of the five top-two gaps was statistically significant. That does not show the top two models are equal, instead it means these task sets cannot actually be reliably separated. On QiskitHumanEval Hard and QCoder, the undirected evidence did lean towards the leader (p = 0.016 and 0.003) before the correction removed it.\n\nThe reality is that these gaps are actually quite small. On QiskitHumanEval Standard, Astra led Gemini 3.1 Pro by 1.1 percentage points. On QuanBench-117, the recorded top two tied at 84.6%. On QuanBench-44, the leader's 95% interval ran from 45.9% to 73.6%. With a few dozen or perhaps a few hundred tasks, scores move that much from sampling alone.\n\n> _\"An observed first place, therefore, should not be read as evidence that the leading model is reliably better than the runner-up.\"_\n\nRankings also shift between suites. Most execution-scored suites agreed  fairly well, but QiskitHumanEval Standard and QCoder rankings agreed at only 0.54. Pick a different benchmark, and you can get a totally different order.\n\n## When the grader is the bug\n\n![Bar chart of GPT-6 Astra's pass@1 on QuantumBenchEval topics T1 (QAOA, 17 tasks) and T6, rescoring the same saved outputs: near 0% with the original grader and above 90% with the corrected grader on both.](https://quantumdev.io/blog/media/image-3e43e165ec194caf50eac5c3207801a6187cb384-1254x826-webp-960.webp)\n\nHere, the researchers did something that not a lot of researchers tend to do. They turned the audit on their own benchmark first. In T1, the evaluator checked exact resource quantities the prompts never specified. In T6, it required a reproducibility value in a form the prompts never stated. Removing only those checks moved Astra to first place among the frontier models on T6, where it had been last. All 32 reference solutions passed under both versions of the grader, and no recorded sample flipped from pass to fail. One caveat from the authors: the rescoring was replayed on a different platform, and 5 of 1,600 samples came out differently. A third topic, T2, also checked undisclosed quantities, but removing those checks changed little, so they report it as recorded.\n\nThen they audited the evaluator of QuanBench, an established benchmark. For QuanBench bench-117, they found 3 defects.\n\n1. Test scripts ran without executing their unittest test cases, so a deliberately wrong answer could pass.\n2. The reference solution ran in the same namespace as the model's code, so an answer could pass by inheriting the referenced function.\n3. A missing dependency imported by the reference solution could make a valid answer fail before it ran.\n\nThe authors showed these mechanisms with controlled probes, and found one false pass in the recorded results. Claude Opus 4.6 was scored as passing QuanBench-117 Task 110 (build a CNOT gate from iSWAP gates), but the circuit it returns is not equivalent to CNOT up to global phase. They are, however, careful about scope. The original evaluator source was not captured at the time, so the probes do not prove that every saved QuanBench outcome came from the inspected version. Their position is that all historical QuanBench scores, passes and failures alike, are recorded outcomes that need to be replayed under an isolated evaluator before anyone reads them as capability estimates. That applies to every QuanBench number on this page.\n\nThey also looked at the 17 QuanBench-117 tasks that no model passed. For each one, the dominant recorded failure traces to an import in the benchmark's own reference or test code, from packages such as `qiskit_algorithms`, `qiskit_optimization`, `qiskit.primitives`, and `qiskit_machine_learning`. That does not mean every model answer on those tasks was correct, since some failed earlier for other reasons. It does mean the apparent shared ceiling was at least partly the benchmark's environment.\n\nLLM judges have their own problem. On QCoder task, QPC004_A2, Two responses that were character-for-character identical and correct (three alternating CNOT gates implementing a swap) were judged as \"PASS\" in one call and \"FAIL\" in another by the same judge model. QCode results carry other caveats too. Its protocol typically shows each model its own earlier attempts, so the five samples are not independent. Its judge changed partway through the study, and the judges are Anthropic models, the same provider as Claude 5.1, which leads QCoder at 43.0% pass@1.\n\n## Empty answers count as wrong answers\n\nGPT-5 returned empty code for 316 of its 755 QiskitHumanEval Standard samples, or 41.9%, and each was scored as a failure. Of those 316, 315 stopped at exactly 2,048 output tokens, which the authors say is consistent with the output budget running out before any code was written. GPT-5's empty rate rose on suites with longer tasks, reaching 63.0% on QCoder.\n\nSo part of GPT-5's score reflects its token limit. On QCoder, GPT-6 Astra, Claude Fable 5.1, and Muse-Glimmer-30B produced no usable code at 2,048 tokens, so their runs used 32,768 tokens instead, a 16x difference. The authors say gaps between those earlier and later QCoder runs mix model capability with output budget and judge version. If you're reading a leaderboard, the output budget is absolutely part of the result.\n\nSyntax is a separate issue. Gemini 3 Flash produced syntactically valid code for only 53% of its QuanBench-117 samples. All 2,650 audited non-empty GPT-6 Astra outputs parsed as Python, and a substantial share still failed the evaluation. Valid Python and correct quantum code are different checks.\n\n## What should you check before relying on a quantum leaderboard?\n\nThe paper ends with six recommendations for benchmark builders. Here they are as questions you can ask any leaderboard or vendor claim.\n\n![Table of six questions to ask of a quantum code leaderboard and why each matters. Do the reference solutions pass, and do the tests run? QuanBench-117's evaluator ran scripts without running their unittest cases. Is the model's code isolated from the reference code? On QuanBench-117 a shared namespace let an empty answer pass a probe. Does the grader check only what the prompt asked for? Undisclosed checks held GPT-6 Astra to 0% on QBE topic T1; removing them gave 94.1%. Are empty or invalid outputs reported? GPT-5's empty-output rate reached 63.0% on QCoder. Are confidence intervals and corrected comparisons shown? No top-two gap on the five primary suites was significant after Holm correction. Were the samples drawn independently? QCoder's sequential protocol means its pass@k is not the standard estimator.](https://quantumdev.io/blog/media/image-51956849fcf2369f82d441c6539ffcb1ddd12188-1232x826-webp-960.webp)\n\nI think that there are two more questions that are worth adding. Which judge model and version graded it? QCoder's judge changed between runs, and no run was rejudged.\n\nIs the suite saturated? A top score on QiskitHumanEval Standard no longer separates the strongest models.\n\n## How do you check LLM-generated Qiskit yourself?\n\n1. Check what the code does.\n\nCode similarity is a weak signal: CodeBLEU correlated with pass/fail at r = 0.280 on QiskitHumanEval Standard. Compare unitaries with Qiskit's `operator.equiv` (equivalence up to global phase), statevectors for state prep, and distributions with a tolerance for sampled outputs.\n\n1. Run it in a clean environment.\n\nSome failures in the paper were imports, such as `UnitaryGate`from `qiskit.quantum_info`, which does not export it. Pin Qiskit and let import errors surface in CI.\n\n1. Make sure your tests run\n\nFeed your harness one deliberately wrong candidate and one empty one. Both should fail.\n\n1. Spell out your contract\n\nIf a test checks quantity, return type, or format, put it in the prompt. T1 and T6 show what happens otherwise.\n\n1. Sample more than once\n\nQuantum-specialized and small open-weight models gained up to 38 points from pass@1 to pass@5 (Granite-Qiskit on recorded Quanbench-117); the strongest models gained about 6 at most. Choosing among samples still needs a real verifier.\n\n1. Watch for truncation\n\nIf a response cuts off mid-block, raise the output budget before you blame the model.\n\n## Where this leaves us\n\nOn short Qiskit API task, the strongest models are near the ceiling of QiskitHumanEval Standard. On QuantumBenchEval, results depend heavily on topic: GPT-6 Astra and Claude Fable 5.1 both reached 94.1% on corrected T1, yet no model exceeded 32% on T4. The authors also list their own limits: QuanBench outcomes await replay, decoding settings and output budgets differ across models, and the roster reflects one point in time.\n\nThe ORNL team says it is releasing its harness, per-sample outcomes, and audit artifacts, along with a public leaderboard called Quantum Arena, so new models can be evaluated under the same protocol. The paper does not give a URL for it.\n\nWhat does your team use to verify AI-generated quantum code today?\n\n## Why should you care?\n\nA leaderboard score is a rough hint about a model, not a final answer on whether it's good, and sometimes the test itself is the thing that's broken. Tests that hand the model a starter template tend to go a lot better than ones where it starts from scratch. Before you pick a tool, try a few models on real tasks from your own project and see which one holds up. Whatever model you use, still read what it writes and run it against tests you trust, because code that looks right and code that works are not always the same thing.\n\nAs you continue to build on AWS, this lands pretty directly on Amazon Braket. Braket has its own Python SDK with its own way of writing circuits, and there's also a Qiskit-Braket provider that lets you run Qiskit circuits on Braket simulators and quantum hardware. Every benchmark in this paper is Qiskit-focused, and the paper did not test Braket at all. My read is that a model's Qiskit leaderboard score tells you very little about how well it writes Braket SDK code. If you're asking an assistant for Braket code, test it on Braket code.\n\nThe nice part is that Braket gives you inexpensive places to do that testing. The Braket SDK comes with a free local simulator that runs on your own machine, and there are managed simulators too, like SV1 for general circuits and DM1 for simulating noise. The managed simulators bill by how long each simulation runs. Real quantum hardware is built differently, with a fee per task plus a fee per shot, and the per-shot price depends on the device. Run AI-written circuits on the local simulator first, move to a managed simulator if you need more qubits or noise, and only spend on hardware once the code has passed your own checks.\n\nThose checks are where the paper's lesson really pays off. \"It ran without errors\" is not the same as \"it's right,\" and the paper found graders that passed wrong answers and failed correct ones. Write a free unit test that checks results you already know should come out, like a Bell pair giving you mostly 00 and 11, and run them on the simulator every time the model changes something. Feed your tests a wrong answer and an empty one once, just to make sure they actually fail.\n\nIf you use Braket's managed Jupyter notebooks or Hybrid Jobs for bigger experiments, keep the same habit: simulator first, real tests, then hardware.\n\nI get it, the paper is all Qiskit. But I'd really encourage you to build on AWS with Amazon Braket and try a small version of these. Test yourself with the Braket SDK. Write a handful of Braket tasks, each with unit tests that check the result on the local simulator. Then give a model each task in two ways, once with a starter template and once without, the same idea as the paper's standard and hard prompts. Run each one several times instead of once, try a couple of different models, and before you trust any of the scores, feed your grader a known-wrong answer and an empty one to make sure both fail. This would be your own small experiment, not a copy of the paper's benchmark, and a handful of tasks won't crown a winner, but it will tell you a lot more about which assistant actually writes good Braket code for you than any Qiskit leaderboard will.\n\nArticle Link: [https://arxiv.org/html/2610.28678#abstract1](https://arxiv.org/html/2610.28678#abstract1)\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3LcD8jriMyeRJ6qPDqsjp9eF3rT/when-ai-writes-quantum-code-who-checks-the-work) on 4 November 2026.\n"
  },
  {
   "id": "post:teaching-ai-to-catch-quantum-errors",
   "type": "post",
   "title": "Teaching AI to Catch Quantum Errors",
   "url": "https://quantumdev.io/blog/teaching-ai-to-catch-quantum-errors/",
   "markdownUrl": "https://quantumdev.io/blog/teaching-ai-to-catch-quantum-errors.md",
   "description": "How AWS trained NVIDIA's AI pre-decoder for quantum error correction on EC2 GPUs, cutting decoding latency 2.4 to 2.9 times, and what it means for builders.",
   "published": "2026-10-30",
   "updated": "2026-10-30",
   "author": "Christian Perez",
   "text": "# Teaching AI to Catch Quantum Errors\n\nBy Christian Perez, published 2026-10-30\n\nCanonical: https://quantumdev.io/blog/teaching-ai-to-catch-quantum-errors/\n\nHow AWS trained NVIDIA's AI pre-decoder for quantum error correction on EC2 GPUs, cutting decoding latency 2.4 to 2.9 times, and what it means for builders.\n\n## The Short Version\n\nThe AWS Quantum Technologies team trained NVIDIA's open-source AI pre-decoder for quantum error correction on five different EC2 GPU instance types then validated it on a P6 instance. At large code sizes the trained model cut decoding latency per round by roughly 2.4 to 2.9 times while also lowering the logical error rate.\n\nI understand that there's a lot of dense vocabulary in that sentence. This article unpacks it so that builders who are new to quantum can actually understand it. Additionally, we'll explain why a GPU benchmark is actually a quantum milestone and we'll lay out where I think this work should go next.\n\nIf you've been following the quantum technology series, including the articles on superposition and entanglement, we'll be adding on top of that. Previous articles covered how qubits work. This article, however, will cover how we keep them from falling apart.\n\n## What is quantum error correction?\n\nQubits are fragile. Heat, stray signals, and imperfect control pulses knock them out of their quantum state, a process called decoherence. On today's hardware that happens faster than most useful algorithms can finish.\n\nQuantum error correction (QEC) is the fix. Instead of trusting one physical qubit, you spread one logical qubit across many physical ones. You cannot simply copy a qubit to make backups so QEC uses a clever trick. It runs a repeating loop:\n\n1. Measure the stabilizers. These are parity checks across groups of qubits. They answer, \"Do these neighbors still agree?\" without reading the actual quantum data so the logical state survives.\n2. Collect the syndrome. Each stabilizer measurement returns a single bit: 0 if its group of qubits still agrees, 1 if something flipped. The full string of those bits is the syndrome. Think of it as smoke detectors going off in certain rooms. You know roughly where the fire is without opening every door.\n3. Decode. The classical computer reads the syndrome, infers the most likely errors, and reports or applies a correction.\n4. Repeat: on superconducting hardware, each round takes about 1 μs.\n\n## How did the team train the decoder?\n\nNewer GPUs trained the pre-decoder up to about 39% faster, but the cheapest path to a good model was not always the newest instance. The setup was quite rigorous. Every run used the same container image on AWS Parallel Computing Service, so hardware was the only variable. This combined CUDA and PyTorch for training, Stim to simulate noisy quantum circuits and generate synthetic syndromes, and the Ising decoding repo for the model itself.\n\nAdditionally, each training sample was one shot of a distance-9 surface code memory experiment: a logical qubit encoded in a 9-by-9 grid of data qubits, measured over 9 rounds, with a physical error rate of 0.006 per operation. Each epoch held about 67 million shots spread across 8 GPUs.\n\nThe practical Playbook is the following:\n\n- Exploration hyperparameter suites: G6e. You get near-H100 speed at a large discount, which matters when you are running dozens of experiments.\n- Final training runs or larger decode distances: P5en or P6, when time between models or GPU memory is the constraint.\n\n![Table comparing five EC2 instance types with eight GPUs each for training the decoder: P4d (NVIDIA A100), the baseline, slowest and oldest generation; G6e (NVIDIA L40S), within about 6% of P5 and about 45% cheaper per hour; P5 (NVIDIA H100), mid-pack and the reference for G6e; P5en (NVIDIA H200), within 1 to 2% of P6 and about 44% cheaper per hour; P6 (NVIDIA B200), fastest, up to about 39% faster than P4d, at a premium.](https://quantumdev.io/blog/media/image-aa7e5997d11e37c5361a20fc37da826fde372718-1262x366-webp-960.webp)\n\n## How did the team validate it?\n\nAt distance 31, the pre-decoder plus MWPM was 2.4x (Fast Variant) to 2.9x (Accurate Variant) faster per round than MWPM alone, and it was more accurate too.\n\nThe team ran validation sweeps on a single B200 GPU, scaling surface code from distance 5 up to distance 31 at the same 0.006 physical error rate. Two findings stand out:\n\n- Small codes favor MWPM alone. Low distances, the fixed overhead of running a neural network costs more than it saves.\n- Large codes favor the pre-decoder. As the syndrome volume grows, MWPM slows down, and clearing the easy errors first pays off in both speed and logical error rate.\n\nWhat we can deduce from this is that the useful fault-tolerant machines will need large code distances, and that is exactly where the AI approach starts to win. The results line up with what NVIDIA reported independently.\n\nOne important nuance from the original post, production decoding has to run on classical hardware physically next to the quantum processor, because of the microsecond deadline. The cloud is not where real-time decoding happens. It is where decoders get built, trained, and improved, and none of that requires a quantum computer.\n\n## Why does this matter for builders?\n\nThe biggest bottleneck in quantum computing right now is partly a classical software problem, and you already have the skills to work on it.\n\n1. You do not need a quantum computer to contribute. Every step in this workflow, from data generation to training to validation, is run on GPUs. Stim simulates the noisy circuits. PyTorch trains the model. If you have trained a model before, you have done 80% of this job.\n2. Quantum progress now depends on machine learning engineers. For years, quantum felt like a physics PhD club. Error correction changes that. The decoder is a pattern recognition problem with a latency budget, which is the kind of problem that the machine learning engineer and systems community solves every day.\n3. This is a real cloud architectural lesson. \"We need a container image, so hardware is the only variable\" is how you run a fair benchmark on anything. Matching instance type to the stage of the work, cheap for exploration and premium for final runs, is a FinOps habit worth building early.\n4. It shows where quantum and AI actually meet. In this study, AI is actually making quantum computers usable by keeping up with their error rate. That is a story you can explain in an interview, a meetup, or a design review.\n\n![Table of six follow-on projects with difficulty and main tools: cost per trained model, a weekend project (EC2 Spot, Trainium, AWS PCS); realistic noise models, intermediate (Stim, Amazon Braket calibration data); generalization sweeps, intermediate (Ising Decoding, AWS PCS); an open decoder benchmark, intermediate and a community effort (GitHub, AWS PCS, containers); an FPGA latency prototype, advanced (EC2 F2, quantization toolchains); new codes and logical operations, research (Stim, custom decoders).](https://quantumdev.io/blog/media/image-145da80e8c1c85283172a74b3fc60b29c6325e54-1256x414-webp-960.webp)\n\n## What doors does this open?\n\nThe cloud becomes the decoder factory, and that creates new roles, new products, and new research paths.\n\n- Decoder development as a cloud workload. Training, sweeping, and characterizing decoders at large scale is compute-hungry and hardware-independent. This is a natural fit for elastic GPU capacity, the same way drug discovery and weather modeling became cloud workloads.\n- Hardware-agnostic iteration. Because the training data is simulated, teams can design decoders for qubit architectures that do not exist yet, then retarget them as the hardware emerges.\n- A new specialization: the QEC machine learning engineer. Someone who understands surface codes, neural network design, and latency budgets at the same time. Very few people sit at that intersection today.\n- Benchmarking as a service. The reproducible PCS workflow is a template. Hardware vendors, labs, startups all need a fair way to compare decoders across accelerators and code sizes.\n- A bridge to fault tolerance. Every percentage point of decoder speed buys headroom for larger codes, and larger codes are the path from noisy experiments to machines that are useful algorithms.\n\n_To learn more, read the blog posted on the AWS Quantum Technologies page:_\n\n\\_\\_[_https://aws.amazon.com/blogs/quantum-computing/decoding-quantum-errors-with-nvidia-ising-decoding-on-aws/_](https://aws.amazon.com/blogs/quantum-computing/decoding-quantum-errors-with-nvidia-ising-decoding-on-aws/)\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3LQQdF7UmyogqTtwfmjWTjHrPTb/teaching-ai-to-catch-quantum-errors) on 30 October 2026.\n"
  },
  {
   "id": "post:quantum-advantage",
   "type": "post",
   "title": "Quantum Advantage",
   "url": "https://quantumdev.io/blog/quantum-advantage/",
   "markdownUrl": "https://quantumdev.io/blog/quantum-advantage.md",
   "description": "What quantum advantage means, which problems to try first, and how AWS builders can start this week, with a fault-tolerant QuEra system coming to Amazon Braket.",
   "published": "2026-09-06",
   "updated": "2026-09-06",
   "author": "Christian Perez",
   "text": "# Quantum Advantage\n\nBy Christian Perez, published 2026-09-06\n\nCanonical: https://quantumdev.io/blog/quantum-advantage/\n\nWhat quantum advantage means, which problems to try first, and how AWS builders can start this week, with a fault-tolerant QuEra system coming to Amazon Braket.\n\nQuantum advantage is the point where a quantum computer, working alongside classical computers, solves a real problem faster, more cheaply, or more accurately than the best classical methods alone. Think of a quantum processor as a specialized accelerator, closer to a GPU than to a new laptop.\n\nOrdinary computers store bits that are either 0 or 1. Quantum computers store qubits, which can hold a blend of both states (0 and 1) at the same time. These qubits can then be linked through entanglement.\n\nFor a narrow class of problems, those properties can cut through work that grows rapidly on ordinary machines.\n\nBut let's start with a short clip to understand why this the topic of Quantum Advantage is so controversial.\n\nWatch the clip on YouTube, from 7:00: [video](https://www.youtube.com/watch?v=IkOicF8qBFU&t=420s)\n\n## Ok but on a technical level, what is Quantum Advantage?\n\nThis is one of the most succinct definitions I've found:\n\n> _Quantum advantage is the execution of a well-specified information-processing task on quantum hardware such that (i) the output can be rigorously validated, and (ii) the task is completed with demonstrably better efficiency, cost, or accuracy than is attainable with the best available classical computation alone._\n\n![Infographic titled 'Quantum Advantage': a quantum computer performs a task no classical computer can in a reasonable time. A chart shows classical solve time rising much faster than quantum as problem size grows, with uses in drug discovery, cybersecurity, materials science and optimization.](https://quantumdev.io/blog/media/image-38590947b7074b103dec53fc0aaf309e917bfd1f-1792x1008-webp-960.webp)\n\nThis is precisely what we are building toward.\n\nWe are building toward a time where quantum advantage is defined by rigor: a well-specified information-processing task executed on quantum hardware whose output can be credibly validated.\n\nWe are building toward demonstrations that clear the real bar by showing demonstrably better efficiency, cost, or accuracy than the best available classical computation alone.\n\nAnd we are building toward trust: results that survive improved classical algorithms, hybrid workflows, and independent verification, so advantage becomes reproducible rather than rhetorical.\n\nSee the post on X: [x.com/i/status/1399738768794529792](https://x.com/i/status/1399738768794529792)\n\nAdvantage is the word that matters for builders. It asks whether anyone should pay to run the job. Candidate claims of useful advantage still get stress-tested by better classical algorithms. Practical, commercially relevant advantage has not been settled. That is not a reason to wait. It is a reason to pick the right problems.\n\n## Got it. Now, what problems should I try to solve?\n\nStart with problems that are already quantum in nature. Molecules, materials, and catalysts are made of electrons whose behavior gets expensive to simulate as systems grow. Here is a one-million dollar term: The Variational Quantum Eigensolver. VQE is the standard method: a classical optimizer proposes circuit parameters, a quantum device estimates an energy, and the loop repeats.\n\nIn plain terms, VQE lets a classical computer and a quantum chip work together to find the lowest-energy state of a molecule without having to simulate every electron exactly. This hybrid approach turns an otherwise intractable quantum chemistry problem into something we can actually run on today’s noisy hardware.\n\nAnalog Hamiltonian simulation on QuEra Aquila is the other strong path. You program the physics you want the hardware to evolve, which fits many-body and materials questions.\n\nImagine you’re trying to understand how a crowd moves through a stadium during an emergency. One way is to write down equations for every person’s speed, direction, and interactions. This is like gate-based quantum computing, where you break everything into discrete operations and simulate step by step. The other way is to build a scaled-down model of the stadium, fill it with real people, and just watch what happens. That’s analog simulation.\n\nOn QuEra’s device, the “people” are ultra cold atoms arranged in a grid, and the “rules of movement” are set by lasers and magnetic fields that you tune. Instead of compiling a long list of quantum gates to mimic a molecule or material, you configure the hardware so its natural behavior is the system you’re studying. Then you let it evolve and measure the outcome directly.\n\nThis approach skips the overhead of translating your problem into gates, which is especially helpful when the system’s own physics already matches what the hardware can do.\n\nFinally, combinatorial optimization is worth learning as a craft, not as a guaranteed win. The Quantum Approximate Optimization Algorithm (QAOA) maps Max-Cut, scheduling, and network design onto a parameterized circuit. Think of QAOA like teaching a robot to solve a puzzle by trial and error: you give it a rough strategy (the circuit), let it tweak its approach (the parameters), and check if it’s getting closer to the best solution. You use problems like Max-Cut or scheduling not because quantum will definitely beat classical methods today, but because they’re perfect practice grounds for learning how to translate real-world problems into quantum circuits, clean up noise, and measure progress.\n\nUse those examples to practice mapping, compilation, error mitigation, and benchmarking against a strong classical solver on EC2 or GPUs. Do not assume a quantum circuit will beat a tuned heuristic on a production instance today. In other words, treat quantum optimization like a training gym: you’re building skills and intuition, not expecting to win a championship match right now. The real value is in mastering the workflow (problem encoding, circuit design, error handling, and fair comparison) so you’re ready when hardware and algorithms mature enough to deliver an actual edge.\n\nAWS is aiming its first [fault-tolerant Braket system, QuEra Libra](https://quantumcomputingreport.com/quera-outlines-2028-roadmap-for-256-logical-qubit-libra-system-and-expanded-aws-cloud-partnership/) around 2028, at chemistry, materials, and high-energy physics.\n\n![Banner reading 'QuEra FTQC System Coming to Amazon Braket in 2028', with the QuEra and Amazon Braket logos.](https://quantumdev.io/blog/media/image-40d8ed8f43c5231623fccc5ca092aca25ab3ff9e-1732x974-webp-960.webp)\n\n## So what can I do this week to get started?\n\nThis week, clone the Amazon Braket examples repo, run a notebook on a simulator, pick a small molecule or graph, run a Hybrid Job, and keep a classical baseline. Start by forking the Braket examples repository to your GitHub, then launch a Jupyter notebook on Amazon Braket and run the “VQE for H₂” or “MaxCut on a 4-node graph” tutorial end-to-end on the local simulator.\n\nOnce that works, pick one concrete problem like the ground-state energy of lithium hydride (LiH) and modify the notebook to run it as a Hybrid Job on Braket’s managed service, which handles the classical-quantum loop for you.\n\nAfter that you can spin up a t3.medium EC2 instance (or use your laptop) to run a classical baseline: for chemistry, use PySCF or Qiskit Nature’s classical eigensolver. Record both results in a shared notebook or markdown file: energy values, cut sizes, runtime, cost, and any errors you hit.\n\nThe goal is to build the reflex of always asking, “What would a classical method give me here, and how does this quantum run compare?”\n\n## Isn't re:Invent 2026 less than 100 days away?\n\nIt absolutely is!\n\nAnd if you're interested in learning more about quantum computing, be sure to start adding some of the quantum computing sessions to your favorites.\n\nTo learn more be sure to check out the following resources:\n\nAmazon Braket Documentation: [https://docs.aws.amazon.com/braket/](https://docs.aws.amazon.com/braket/)\n\nThe Quantum Environment: [https://quantumenv.dev](https://quantumenv.dev)\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3IyP2NEZSwORnSyG3fibaj9Rfb7/quantum-advantage) on 6 September 2026.\n"
  },
  {
   "id": "post:navigating-quantum-innovation",
   "type": "post",
   "title": "Navigating Quantum Innovation",
   "url": "https://quantumdev.io/blog/navigating-quantum-innovation/",
   "markdownUrl": "https://quantumdev.io/blog/navigating-quantum-innovation.md",
   "description": "Michael Brett of Amazon Braket on what is easy and hard in quantum computing, and where AWS builders should focus: algorithms, benchmarks and AI.",
   "published": "2026-08-19",
   "updated": "2026-08-19",
   "author": "Christian Perez",
   "text": "# Navigating Quantum Innovation\n\nBy Christian Perez, published 2026-08-19\n\nCanonical: https://quantumdev.io/blog/navigating-quantum-innovation/\n\nMichael Brett of Amazon Braket on what is easy and hard in quantum computing, and where AWS builders should focus: algorithms, benchmarks and AI.\n\nAt the core of technological innovation lies a fundamental exercise in problem-set triage: identifying what is easy and what is hard. In software engineering, these answers tend to dictate the trajectory of our projects. The projects we undertake result in bad builds, good builds, and truly transformative architectures. What we build provides lessons learned. These lessons drive fields forward by making processes more efficient or uniquely effective.\n\nAt the core of this innovation is the question we started with: what are the easy and hard problems to solve?\n\nSolving difficult problems is where the highest return on investment exists. As developers and innovators aim to move the needle in quantum computing, these difficult problems are exactly where efforts must be concentrated.\n\nMichael Brett, Head of Go-To-Market for Amazon Braket on AWS, articulated this precise dynamic on \"Beyond the Quantum Headlines\", a podcast by Resonance. The discussion distilled the current state of quantum computing and provided a clear North Star for where AWS builders should focus their attention to truly drive innovation.\n\n![Thumbnail of the video 'When Quantum Meets AI' from Beyond the Quantum Headlines, showing four speakers.](https://quantumdev.io/blog/media/image-660e99a21f7c6b8099336d561806f243e22a20bd-760x394-webp-640.webp)\n_YouTube video at the end of the article._\n\nI would like to take this opportunity relay what I learned, and where you can start focusing your efforts.\n\n## What is easy and what is hard in quantum computing?\n\nTo build effectively on AWS, developers must first understand the current operational landscape of quantum computing. The industry has matured enough that certain historical barriers are now trivial, while other, more complex challenges have surfaced. Michael identified two easy problems and three hard problems.\n\n### The Easy Problems\n\n### Accessing a Quantum Computer\n\nHistorically, accessing quantum hardware required millions of dollars, specialized cryogenic facilities, and deep institutional partnerships. Today, access is functionally a solved problem. Amazon Braket acts as a fully managed broker, eliminating the astronomical overhead of hardware procurement. Through a single API, developers can access a variety of Quantum Processing Units (QPUs) from providers like IonQ, Rigetti, and QuEra.\n\n### Programming a Quantum Computer\n\nLike standard computer code, writing quantum code has similarly been democratized. AWS provides fully managed Jupyter notebook environments pre-configured with the Amazon Braket Python SDK and integration with popular libraries like PennyLane. Developers do not need to learn proprietary languages for each hardware provider. They can build circuits using standard gates, test them on managed classical simulators, and execute them on physical hardware by changing a single line of code.\n\n### The Hard Problems\n\nThe frontier of quantum computing is in implementation and optimization. These are the areas where real subject matter expertise is required.\n\n## How do you match algorithms to quantum hardware?\n\nSelecting an algorithm is only the beginning. Developers need to map their specific problem set to the correct quantum algorithm and then optimize that algorithm to handle the constraints of modern quantum hardware. Quantum states are highly fragile. Algorithms must be designed to execute within the limited coherence time of the qubits while minimizing gate depth to reduce accumulated error before the system decoheres.\n\n### Fitting Algorithms to Target Systems\n\nQuantum hardware is built differently by different companies. Different providers use entirely different physical modalities, such as trapped ions, superconducting circuits, or neutral atoms. Each system has unique characteristics like topology, connectivity graphs, and native gate sets. A circuit optimized for a superconducting processor will likely perform poorly on a trapped-ion system if it is not heavily adapted. Developers need to understand the hardware they are targeting and compile their algorithms to use specific physical architectures efficiently.\n\n### Optimizing Overall Runtime\n\nQuantum computers do not operate in a vacuum. Most practical quantum algorithms today are hybrid, requiring tight feedback loops between classical high-performance computing resources and quantum processors. Optimizing the overall runtime of these hybrid workloads is a profound engineering challenge. Data transfer latency, classical computation time, and quantum execution queueing must all be orchestrated seamlessly to prevent systemic bottlenecks.\n\n## How do quantum computing and AI help each other?\n\nThe method by which quantum computing and artificial intelligence work together is a key area of current research, but that dynamic flows in two opposing directions. Here, Michael discussed the two-way relationship that exists between Quantum and AI: how quantum augments AI and vice-versa, how AI augments quantum.\n\n### Quantum for AI\n\nUsing quantum computers to accelerate artificial intelligence workloads still contains a lot of unknowns. While the theoretical potential is there, there is no definitive, proven methodology for integrating quantum processing into standard large language model training or inference pipelines.\n\n### AI for Quantum\n\nOn the other hand, the application of artificial intelligence to advance quantum computing is a resounding success. Machine learning models are routinely deployed to decode error syndromes in quantum error correction. Neural networks optimize circuit compilation by discovering more efficient ways to route qubits across physical hardware grids. AI is enabling everyone from enterprise developers to academic researchers to innovate faster within the quantum field. It acts as a force multiplier, automating the complex classical calculations required to keep quantum systems stable.\n\n## What matters in quantum benchmarking?\n\nAs new quantum processors enter the market, evaluating their utility requires objective metrics. Specifications like qubit count or coherence time are helpful, but they don't capture the overall performance of a system on real enterprise workloads.\n\n### Unitary Foundation's Metriq\n\nTo address this, the industry relies on platforms like Metriq, which, according to Michael, the team at Amazon Braket find great value in. Metriq is maintained by the Unitary Foundation. Metriq is an open-source, community-driven benchmarking platform. It aggregates benchmark executions from system-level hardware characterization to application-inspired tasks like quantum machine learning and optimization. This ultimately creates a scoreboard that allows developers to compare cross-platform performance reliably.\n\n### The Customer Reality\n\nFor enterprise developers and business leaders, however, the only benchmark that truly matters is the price-to-performance ratio compared to a purely classical computing alternative. Understanding this reality is crucial for having the proper scope. The current goal is achieving quantum readiness, exploring algorithmic suitability, and building hybrid architectures that will scale once true quantum advantage is realized in the hardware.\n\n## Where is AWS investing in quantum computing?\n\nThe most helpful insight I gathered from the podcast was recognizing where AWS is heavily investing. Given that the hard problems are in orchestration and runtime optimization, AWS is aggressively investing in high-performance compute integration for Amazon Braket.\n\nThe feature driving this effort is Amazon Braket Hybrid Jobs. This service offers a fully managed way to run hybrid quantum-classical algorithms that require both classical AWS compute instances and physical quantum processing units.\n\nInstead of asking developers to manually provision servers, configure network routing, and manage quantum execution queues, Hybrid Jobs spins up the requested classical compute, executes the classical logic, triggers the quantum tasks, and shuts down the instances upon completion. Furthermore, tasks submitted through Hybrid Jobs benefit from priority queueing on the target Quantum Processing Unit. This priority access is ideal for iterative algorithms like the Quantum Approximate Optimization Algorithm or Variational Quantum Eigensolver. In these algorithms, the classical system must continuously update parameters based on immediate quantum results, and normal queue wait times would make this process too slow.\n\nAWS further simplifies the developer experience by integrating Hybrid Jobs with Amazon CloudWatch and Amazon S3. Developers can monitor algorithm progress in near-real time. This enables them to track custom metrics like training loss or expectation values, while all finalized results are uploaded into their designated S3 buckets.\n\n### Automating Complexity with Amazon Bedrock\n\nAs hybrid quantum and classical workloads grow more complex, AWS is actively investing at the intersection of generative AI and quantum computing by working to seamlessly integrate Amazon Bedrock with Amazon Braket. This convergence provides builders with a new toolkit to accelerate their development cycles. Developers should innovate ways to deploy foundation models directly into Braket managed JupyterLab environments. This would effectively lower the barrier to entry for writing quantum circuits. Bedrock agents can also be utilized to streamline the hybrid architecture. By automating the translation of classical data into quantum inputs, such as mapping complex network topologies to node arrangements for optimization tasks, generative AI removes a lot of the classical preprocessing bottlenecks. For developers looking to start building with quantum today, connecting Bedrock with Braket can act as a force multiplier. The goal is to abstract away operational friction, allowing engineering teams to get past the setup phases and focus their time and capital on algorithm design and innovation.\n\nBy abstracting away the heavy lifting of infrastructure management, AWS allows developers to focus their engineering capital exactly where it belongs. They can stop worrying about how to access the hardware and start solving the hard problems of algorithm optimization, effectively pushing the boundaries of quantum innovation.\n\nIf you are interested in tackling difficult problems, building on AWS, and innovating with quantum computing, I highly encourage you to watch the podcast, come to your own conclusions, and decide where you want to innovate at the bleeding edge of AI and Quantum Computing.\n\nWatch the episode on YouTube: [When Quantum Meets AI](https://www.youtube.com/watch?v=It8jBaUpBn4)\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3I10NDOsUMD2mtqY7sIfD6oKb6s/navigating-quantum-innovation) on 19 August 2026.\n"
  },
  {
   "id": "post:entanglement-on-amazon-braket-the-cnot-gate-and-your-first-bell-state",
   "type": "post",
   "title": "Entanglement on Amazon Braket: The CNOT Gate and Your First Bell State",
   "url": "https://quantumdev.io/blog/entanglement-on-amazon-braket-the-cnot-gate-and-your-first-bell-state/",
   "markdownUrl": "https://quantumdev.io/blog/entanglement-on-amazon-braket-the-cnot-gate-and-your-first-bell-state.md",
   "description": "A hands-on guide to quantum entanglement on Amazon Braket. Build a Bell state with the CNOT gate, run it in 2 lines, and measure the correlation.",
   "published": "2026-07-19",
   "updated": "2026-07-19",
   "author": "Christian Perez",
   "text": "# Entanglement on Amazon Braket: The CNOT Gate and Your First Bell State\n\nBy Christian Perez, published 2026-07-19\n\nCanonical: https://quantumdev.io/blog/entanglement-on-amazon-braket-the-cnot-gate-and-your-first-bell-state/\n\nA hands-on guide to quantum entanglement on Amazon Braket. Build a Bell state with the CNOT gate, run it in 2 lines, and measure the correlation.\n\nIn the previous article, we put a single qubit into superposition ([Unlocking Superposition: The Hadamard Gate on Amazon Braket](/blog/unlocking-superposition-the-hadamard-gate-on-amazon-braket/)).\n\nA single qubit in superposition is a good start, but by itself it behaves a lot like a very expensive spinning coin. The real power of quantum computing shows up when qubits start to interact. When 2 qubits become linked so that measuring one tells you something about the other. This is called \"entanglement\" and this is the property that enables a quantum computer to perform.\n\nTo create a state of entanglement between two qubits, we need a new tool: the controlled-NOT gate, usually written as CNOT. By the end of this article you will understand what entanglement is, what it is not, and you will have written and run a real entangling circuit on Amazon Braket.\n\nAs with the last article, some of the vocabulary here can feel foreign at first. I have linked references from the [Quantum Learner glossary](https://quantum.altivum.ai/glossary) throughout so you can go deeper on any concept that catches your attention.\n\n_A quick recap of the last article:_\n\n- _A qubit is the quantum version of a classical bit, and unlike a bit it can exist in a superposition of 0 and 1 at the same time._\n- _The Bloch sphere is a globe that represents every possible state of a single qubit, with 0 at the North Pole and 1 at the South Pole._\n- _Because a qubit in superposition is probabilistic, you run a circuit many times to see the statistics. Each run is called a shot._\n\nSuperposition gives you multiple possibilities out of one qubit. That alone does not beat a classical computer. The leap happens when you let qubits influence each other, and that requires a gate that acts on more than one qubit at a time.\n\n## What is entanglement?\n\nEntanglement is a shared state between qubits. When 2 or more qubits are entangled, they are described by a single combined state. The pair stops being \"qubit A doing its thing and qubit B doing its thing\". Instead, it becomes a single object with correlated outcomes. The implication here is that if you measure one qubit in an entangled pair, you immediately learn something about the other, no matter how the 2 are arranged. Their results are locked together.\n\nA simple mental image is tends to be very helpful. Imagine 2 coins that are prepared together in a special way where what one coin lands on immediately tells you what the second coin lands on. Each individual flip of a single coin lands randomly, so you cannot predict any single result but seeing what one coin lands on (measuring) tells you exactly what the second coin landed on without having to look at it. Across many flips, the 2 coins always agree with each other. That agreement comes from how the pair was prepared as a single system. The two coins are entangled.\n\nTo describe 2 qubits, we list the 4 possible combined outcomes:\n\n|00⟩ |01⟩ |10⟩ |11⟩\n\n> _Note: this type of notation is know as Dirac notation. When you see \"|x⟩\" it is pronounced \"ket 'x'\". When you see \"⟨x|\", it is pronounced \"bra 'x'\". Not only do you now know how to speak in Dirac notation but you also know why Amazon named their quantum computing service \"Braket\"._\n\nA separable pair can be written as 2 independent qubits multiplied together. An entangled pair lives in a blend of those 4 states that refuses to factor into independent parts. Holding that idea in mind is enough to follow everything below.\n\n## What Entanglement Is Not\n\nEntanglement attracts more myths than almost any topic in physics. Three of them are worth clearing up before we touch code.\n\n1\\. Entanglement does not send messages faster than light\n\nThis is a very common misconception. When you measure a qubit, you get a random result that you cannot control. A colleague measuring the other qubit also gets a random result they cannot control. Neither of you can force a particular outcome.\n\nThe correlation only becomes visible when you bring the 2 sets of results together and compare them. Comparing requires a normal classical channel, such as a phone call, an email, or a packet sent over the internet, and that channel is limited by the speed of light. This is called the \"no-communication theorem\". Entanglement produces shared correlations, and it gives neither party a way to transmit a chosen message.\n\n2\\. Entanglement is more than a \"pair of gloves in 2 boxes\"\n\nA common way to downplay entanglement is the glove analogy. Put a left glove in one box and a right glove in another, ship them to different cities, and open one box. You instantly know what is in the other. That is ordinary correlation, and there is no quantum mystery in it at all. If entanglement were only this, it would not be worth a series of articles.\n\nThe difference appears in experiments built around Bell's theorem and the CHSH inequality. Quantum correlations violate a limit that every glove-style explanation must obey. In plain language, the measurement outcomes cannot be accounted for by each qubit secretly carrying a pre-set answer chosen at preparation time. The correlations are stronger than any pre-agreed classical strategy could ever produce. You do not need to prove this to build with it, but it is the reason entanglement counts as a genuine computational resource that will soon bring forth quantum advantages.\n\n3\\. Measuring one qubit does not push the other where you want\n\nMeasurement reveals the shared state. It does not hand you a lever to force the far qubit into a value of your choosing. The randomness sits at the foundation of the physics, which is exactly why the faster-than-light shortcut in the first myth does not work.\n\n## What does the CNOT gate do?\n\nThe controlled-NOT gate is a 2-qubit gate with a control qubit and a target qubit. Its rule is short:\n\n- If the control qubit is 1, flip the target qubit.\n- If the control qubit is 0, leave the target qubit alone.\n\nLike every quantum gate, CNOT is reversible. Its matrix is a 4x4 unitary, and applying it twice brings you back to where you started. The diagram below shows the standard notation.\n\n![Circuit diagram of a CNOT gate: a solid dot on the control line, joined by a vertical line to a circled plus on the target line.](https://quantumdev.io/blog/media/image-39a1fddb9c92bee892c3c534ec7d68d896bd608a-1312x670-webp-960.webp)\n_The solid dot marks the control qubit. The circle with the cross marks the target. The target flips only when the control is 1._\n\nThe full behavior fits in a truth table:\n\n| Control (in) | Target (in) | Control (out) | Target (out) |\n| --- | --- | --- | --- |\n| 0 | 0 | 0 | 0 |\n| 0 | 1 | 0 | 1 |\n| 1 | 0 | 1 | **1** |\n| 1 | 1 | 1 | **0** |\n\nLook closely at the bottom 2 rows. When the control is 1, the target value changes. When the control is 0, nothing happens to the target.\n\nNow, if you run CNOT on plain 0s and 1s, all it does is copy or flip bits. Feed it definite classical inputs and you get definite classical outputs, with no entanglement anywhere. The interesting behavior only appears when the control qubit is itself in superposition. That is why the Hadamard gate from the last article is so incredibly important.\n\n## How does a CNOT gate make a Bell state?\n\nEntanglement comes from combining the 2 gates you now know:\n\n1. Apply a Hadamard gate to qubit 0 to put the control into superposition.\n2. Apply a CNOT gate with qubit 0 as the control and qubit 1 as the target.\n\nFollow the state as it moves through the circuit:\n\n- Start with both qubits at 0. The system is |00⟩.\n- Apply H to qubit 0. Qubit 0 spreads into a superposition, so the pair becomes an equal mix of |00⟩ and |10⟩, written as (|00⟩ + |10⟩) / √2.\n- Apply CNOT. It flips qubit 1 wherever qubit 0 is 1. The |00⟩ term is left alone, and the |10⟩ term turns into |11⟩. The result is (|00⟩ + |11⟩) / √2.\n\nThat final state is a Bell state, the simplest and most famous entangled state in all of quantum computing.\n\n![Circuit diagram of a Bell state: two qubits start in |0⟩, an H gate acts on qubit 0, a CNOT joins qubit 0 to qubit 1, and both are measured, giving (|00⟩ + |11⟩)/√2.](https://quantumdev.io/blog/media/image-ceeffb7281c61ab4126bf1aefe8c92683d51d792-1464x714-webp-960.webp)\n_A Hadamard on qubit 0 creates superposition, and the CNOT ties qubit 1 to it. The 2 measurement boxes on the right read out the result._\n\n## How do you write the circuit on Amazon Braket?\n\nThe workload is 2 method calls in the Braket SDK. Gates are first-class objects, and you chain them onto a circuit.\n\n```python\nfrom braket.circuits import Circuit\n \n# Put qubit 0 in superposition, then entangle qubit 1 with a CNOT\ncircuit = Circuit().h(0).cnot(0, 1)\n \nprint(circuit)\n```\n\nThe `.h(0)` call applies a Hadamard gate to qubit 0. The `.cnot(0, 1)` call applies a CNOT with qubit 0 as the control and qubit 1 as the target. Chaining them defines the whole Bell state circuit in a single line.\n\nBecause the outcome is probabilistic, you run many shots to see the pattern. Here is the circuit on the free Amazon Braket local simulator, which runs entirely on your own machine.\n\n```python\nfrom braket.devices import LocalSimulator\nfrom braket.circuits import Circuit\n \n# 1. Define the local simulator\ndevice = LocalSimulator()\n \n# 2. Build the Bell state circuit\ncircuit = Circuit().h(0).cnot(0, 1)\n \n# 3. Run 1000 shots\ntask = device.run(circuit, shots=1000)\n \n# 4. Fetch and print the measurement counts\nresult = task.result()\nprint(result.measurement_counts)\n```\n\nWhen you print the measurement counts, you see something close to `{'00': 502, '11': 498}`. 2 outcomes, each near 50%. Now look at what is missing. The results `01` and `10` never appear.\n\n## Why do the missing outcomes matter?\n\nThe empty bars carry the meaning of the entire experiment. In a Bell state the 2 qubits always agree, so the outcomes `01` and `10` have zero probability. The data you get back is the correlation itself.\n\nTo see how special that is, compare it against a pair of qubits that are not entangled. Send each qubit through its own Hadamard gate:\n\n```python\n# Two independent superpositions, no entanglement\ncircuit = Circuit().h(0).h(1)\n```\n\nRun 1,000 shots of that circuit and all 4 outcomes show up, each near 250:\n\n```text\n{'00': 249, '01': 253, '10': 248, '11': 250}\n```\n\nEvery combination appears because the qubits pay no attention to each other. Put the 2 circuits side by side and the contrast is clear:\n\n- The entangled circuit narrows 4 possible outcomes down to 2 that always match.\n- The independent circuit spreads evenly across all 4 outcomes.\n\nThat narrowing to correlated outcomes is measurable entanglement. You can watch it happen in the count with no advanced hardware and no cost.\n\n## How do you move from the simulator to a real QPU?\n\nThe path to real quantum hardware is the same swap you learned last time. Replace `LocalSimulator()` with an `AwsDevice` pointed at a specific QPU, and the exact same circuit runs on physical qubits.\n\n```python\nfrom braket.aws import AwsDevice\nfrom braket.circuits import Circuit\n \n# Gate-based trapped-ion QPU from IonQ. Check the Braket console\n# Devices page for the current list of available hardware.\ndevice = AwsDevice(\"arn:aws:braket:us-east-1::device/qpu/ionq/Forte-Enterprise-1\")\n \ncircuit = Circuit().h(0).cnot(0, 1)\ntask = device.run(circuit, shots=1000)\n \nresult = task.result()\nprint(result.measurement_counts)\n```\n\nResults from real devices are stored in your S3 bucket and returned through the same `task.result()` call. Amazon Braket gives you gate-based QPUs from providers such as IonQ, Rigetti, and IQM, and the available devices change over time, so the console is the source of truth for what is live today.\n\nOne honest caveat about physical hardware. On a real QPU you will see a small number of `01` and `10` results that the simulator never produces. Those stray counts are what we refer to as noise, the fingerprint of every quantum computer built so far. Today's devices are not perfect, and learning to read that error is a skill of its own. It is a subject I will return to later in this series.\n\n## Why does this matter for builders?\n\nEntanglement is the resource behind essentially every quantum algorithm that outruns a classical one.\n\n- Quantum teleportation and superdense coding are built directly on entangled pairs.\n- Quantum error correction, the thing that will eventually make large QPUs reliable, depends on entangling many qubits together.\n- The famous speedups from Grover's search and Shor's factoring come from entangling a register of qubits and then using interference to concentrate probability on the right answer.\n\nStrip entanglement out and a quantum computer is just an expensive random number generator. Enable entanglement and the machine can represent and process correlations that a classical CPU cannot reach. For an AWS builder, going from a Hadamard gate to a Hadamard plus a CNOT is the moment Amazon Braket starts being hardware that does something fundamentally new.\n\n## Key Takeaways\n\n- **Entanglement is a shared state.** 2 or more qubits are described by one combined state that cannot be split into independent parts. Measure one and you learn about the others.\n- **The CNOT gate is the entangler.** It flips a target qubit only when the control qubit is 1. On plain classical inputs it merely copies or flips, but paired with a control in superposition it creates entanglement.\n- **H then CNOT builds a Bell state.** The 2-line recipe produces (|00⟩ + |11⟩) / √2, the simplest entangled state and the backbone of countless algorithms.\n- **The proof is in the missing outcomes.** Run 1,000 shots of a Bell state and you get only 00 and 11. The absence of 01 and 10 is entanglement you can measure.\n- **Entanglement is not faster-than-light communication.** You cannot send a chosen message with it. The correlation only appears once results are compared over a normal classical channel.\n- **The same code reaches real hardware.** Swap the local simulator for an AwsDevice ARN and your Bell state runs on a physical QPU, noise and all.\n\n## What comes next\n\nYou can now put qubits into superposition and lock them together with entanglement. The next question is how to steer results on purpose. That is the job of interference, where quantum states add and cancel to push probability toward the answer you want.\n\nIn the next article we will use interference to build a circuit that actually computes something, and we will start looking at how noise on real hardware shapes what you get back.\n\n_For runnable notebooks on Bell states and much more, sign up for_ [_Quantum Learner_](https://quantumlearner.dev) _and start building on Amazon Braket._\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3GbA88OzNzzA2I9mczuxAfNbm2K/entanglement-on-amazon-braket-the-cnot-gate-and-your-first-bell-state) on 19 July 2026.\n"
  },
  {
   "id": "post:a-developers-field-guide-to-amazon-braket",
   "type": "post",
   "title": "A Developer's Field Guide to Amazon Braket",
   "url": "https://quantumdev.io/blog/a-developers-field-guide-to-amazon-braket/",
   "markdownUrl": "https://quantumdev.io/blog/a-developers-field-guide-to-amazon-braket.md",
   "description": "A guided tour of the Amazon Braket console, page by page. What each button does, when you'd use it, and how to run your first task on a real quantum computer.",
   "published": "2026-07-17",
   "updated": "2026-07-17",
   "author": "Christian Perez",
   "text": "# A Developer's Field Guide to Amazon Braket\n\nBy Christian Perez, published 2026-07-17\n\nCanonical: https://quantumdev.io/blog/a-developers-field-guide-to-amazon-braket/\n\nA guided tour of the Amazon Braket console, page by page. What each button does, when you'd use it, and how to run your first task on a real quantum computer.\n\n## Who is this guide for, and how should you read it?\n\nIf you have an AWS account and a little curiosity about quantum computing, you already have everything you need to run a program on a real quantum processor right now. You don't need a physics degree, you don't need to own a dilution refrigerator, and you don't need to email a national lab. All you need is your browser, the AWS Console, and a few minutes to explore.\n\nThis guide is a click-by-click walk through the **Amazon Braket** console. Amazon Braket is the fully managed quantum computing service on AWS. I'm going to open every panel, press every button, and tell you what happens and why it matters.\n\nWhere helpful, I'll drop in a few lines of Python from the Braket SDK so you can go from looking at the console to \"my circuit just ran on trapped ions in a lab in Europe.\"\n\nA few conventions:\n\n- **Screenshots** mark each stop on the tour. Every image caption describes exactly what you're looking at, so you can follow along in your own console.\n- `Try it` callouts are the \"put the guide down and go do this\" moments.\n- **Tables** are your cheat sheets (device specs, pricing, instance types). Skim them, then come back.\n- Everything here reflects the Braket console as of **mid-2026**. Quantum hardware moves _fast_ — devices come and go monthly (there's a whole section on that), so treat specific device names as a snapshot.\n\n> **A note on cost before you touch anything:** Braket has no flat subscription. Simulators and quantum processors are pay-as-you-go, and the local simulator that ships with the SDK is **free**. You can run this entire tour's \"hello world\" for pennies and I'll show you how to set a hard spending limit so you never get surprised. Read the [Cost controls](#10-cost-controls) section before you submit to real hardware.\n\n## How does Braket work, in 30 seconds?\n\nBraket wraps three things behind one console and one SDK:\n\n1. **Simulators** — classical computers pretending to be quantum ones. Perfect for development. Some run locally (free), some run on-demand in the cloud.\n2. **QPUs (Quantum Processing Units)** — this is the real thing. Trapped ions, superconducting circuits, neutral atoms from several hardware companies, all reachable through the same API.\n3. **What brings it all together** — managed Jupyter notebooks, hybrid quantum-classical job orchestration, result storage in S3, cost tracking, and access controls.\n\nThe workflow Braket is built around is a simple loop: **Build → Test → Run → Analyze.**\n\n![Diagram of the Braket loop: Build (design your circuit), Test (run on a simulator), Run (submit to a QPU) and Analyze (read results from S3), with an arrow labelled iterate leading from Analyze back to Build.](https://quantumdev.io/blog/media/image-36b85961edab9ae5cc7ca0c1cccfb48fe4cdbe11-1150x208-webp-960.webp)\n\nYou build and test for virtually $0 and you only pay real money at the \"Run on a QPU\" step. Keep that loop in mind. The whole console is organized around it.\n\n![The Amazon Braket landing page in the AWS console, with a How it works diagram of Build, Test, Run and Analyze, and panels for getting started, Quantum Embark, pricing and more resources.](https://quantumdev.io/blog/media/image-6b397abb4bef14b430a5027587202c013e87ca9f-2974x1808-webp-960.webp)\n\n## 1. Getting In: Finding Braket and Flipping On the Switch\n\n### Finding the service\n\nOpen the AWS Console and type **Braket** in the top search bar. Amazon Braket shows up under **Services**, described as a \"Service for exploring, evaluating, and experimenting with quantum computing.\"\n\nThe search panel also surfaces three Braket **features** directly (_Braket Direct_, _Permissions and Settings_, and _Service Dashboard)_ so you can jump straight to a subsection. Click the star to pin Braket to your console favorites; you'll be back.\n\n![The AWS console search for Braket, with Amazon Braket first among the services, and Braket Direct, Permissions and Settings and Service Dashboard listed as features.](https://quantumdev.io/blog/media/image-fd953fd75eb7d2286ec8654c11414a888fb70163-1876x1636-webp-960.webp)\n\n### The left-nav: your whole map\n\nOnce inside, everything hangs off the left navigation menu. Here's the entire console in one table. This is also the map for the rest of the guide.\n\n| Nav item | What it's for | Covered in |\n| --- | --- | --- |\n| **Dashboard** | At-a-glance widgets: notebooks, jobs, announcements, preferred device | Section 2 |\n| **Devices** | Browse every simulator and QPU; inspect specs, calibration, cost | Section 3 |\n| **Notebooks** | Managed JupyterLab instances, pre-loaded with the SDK | Section 4 |\n| **Hybrid jobs** | Orchestrated quantum-classical workloads with priority device access | Section 5 |\n| **Quantum tasks** | The atomic unit of work — one circuit + shots submitted to one device | Section 6 |\n| **Algorithm library** | 11 ready-to-run textbook algorithms in Python | Section 7 |\n| **Braket Direct** | Reservations, expert office hours, experimental hardware features | Section 8 |\n| **Administration → Permissions and settings** | Enable third-party devices; manage IAM execution roles | Section 1 & 10 |\n| **Administration → Spending limits** | Hard dollar caps per QPU | Section 10 |\n| **Announcements** | New devices, retirements, feature launches | Section 11 |\n\n![The Amazon Braket left-hand navigation: Dashboard, Devices, Notebooks, Hybrid jobs, Quantum tasks, Algorithm library, Braket Direct, Administration with Permissions and settings and Spending limits, and Announcements.](https://quantumdev.io/blog/media/image-6caa698e4186c7277897e495f76432e9b19afcc8-2218x1150-webp-960.webp)\n\n### Enable third-party devices: the one-time setup\n\nBefore you can send a workload to any non-AWS device (IonQ, IQM, QuEra, Rigetti, AQT), you have to accept the third-party hardware terms once. Go to:\n\n**Administration → Permissions and settings**.\n\nOn the **General** tab you'll find **Enable third-party devices**. Flipping this to _Enabled_ acknowledges that when you use a partner's hardware, your circuit content is transferred to that provider outside AWS facilities for processing. No acceptance, no QPU access.\n\n![The General tab of Permissions and settings, where Enable third-party devices shows Enabled above the terms for sending circuits to third-party hardware providers.](https://quantumdev.io/blog/media/image-fc7beceb917672910650658252845fc70426fdd9-3426x1408-webp-960.webp)\n\n### Execution Roles (the IAM plumbing)\n\nThe **Execution roles** tab is where Braket confirms it has the IAM roles it needs. There are three to know about:\n\n- **Service-linked role** (`AWSServiceRoleForAmazonBraket`) — lets Braket act on your behalf. Braket creates this automatically; you'll see a green \"Service-linked role found\" banner.\n- **Notebook access role** — an IAM role with the `AmazonBraketFullAccess` policy so notebook instances can call Braket operations. Use **Verify existing roles** to confirm you have one.\n- **Hybrid jobs execution role** — carries the `AmazonBraketJobsExecutionPolicy`, the minimum permissions a hybrid job needs to run.\n\nIf any are missing, the page offers a one-click **Create default role**. For a first-time account, click through all three and you're provisioned.\n\n![The Execution roles tab of Permissions and settings: the service-linked role is found, and the notebook access role and hybrid jobs execution role each offer Verify existing roles and Create default role.](https://quantumdev.io/blog/media/image-23eac79434c161bce55720b7bb0ad4cca0e9be98-3432x1350-webp-960.webp)\n\n> `Try it` — In your AWS account, spend your first two minutes here: enable third-party devices, then verify all three execution roles. It's the only mandatory setup, and skipping it is the #1 reason a first QPU submission fails.\n\n## 2. The Service Dashboard\n\nClick **Dashboard** for the mission-control view. It's a grid of widgets you can rearrange, remove, or add to with **Add widgets** (and **Reset to default layout** if you make a mess):\n\n- **Active notebooks** and **Recent hybrid jobs** — quick status, with \"Create\" shortcuts when empty.\n- **Announcements** — the live feed of device launches and retirements (more in [§11](#11-announcements-how-to-stay-current)).\n- **Community support** — direct links to the AWS Quantum Computing Blog, the Quantum Computing Stack Exchange, and AWS re:Post.\n- **Device** — pin a **preferred device** so its live properties surface right here on the dashboard.\n- **Resources** — links to Quantum Embark (AWS's paid advisory program for Enterprise), the Braket Learning Plan, and the developer guide.\n\n![The Braket service dashboard, with widgets for active notebooks, recent hybrid jobs, announcements, community support, a preferred device and resources.](https://quantumdev.io/blog/media/image-ae756e0a7dcf67fef89e79c36696cfffe0c00bff-3420x1776-webp-960.webp)\n\n> Try it — Hit **Set a preferred device** and pick a simulator like SV1. Now your dashboard always shows you a live device you can reach for free.\n\n## 3. Devices: the Heart of Braket\n\nThis is one of the pages you'll open most. **Devices** lists every simulator and QPU you can reach, with provider, device name, live status, and a one-line description.\n\n### The current fleet\n\nOut of the box you see the **online** devices. As of this writing that's nine — two AWS simulators and seven QPUs spanning four qubit technologies:\n\n![The Devices page listing nine devices: the SV1 and DM1 simulators from AWS and QPUs from AQT, IonQ, IQM, QuEra and Rigetti, all online except IonQ Forte Enterprise 1.](https://quantumdev.io/blog/media/image-f7acbbcd1de633021c5d91a69bd5df54b61180d4-3430x1208-webp-960.webp)\n\nThree status values are very important: **ONLINE** (accepting tasks now though QPUs still have availability windows), **OFFLINE** (temporarily unavailable), and **RETIRED** (gone for good).\n\n### The fleet moves fast: \"Show retired devices\"\n\nToggle **Show retired devices** and the list jumps from 9 to 28. This is the single best lesson the Devices page teaches: _quantum hardware is a moving target._ The retired list is a graveyard of an entire industry's iteration.\n\n![The Devices page with Show retired devices switched on, listing 28 devices, most of them retired, among them D-Wave annealers, Rigetti Aspen machines, IonQ Harmony and Aria, OQC Lucy, Xanadu Borealis and the TN1 simulator.](https://quantumdev.io/blog/media/image-92325b657825deeac6692ea2cbcc9f29480e9059-3430x1708-webp-960.webp)\n\nThe takeaway for anyone building on Braket: **program against device _properties_, not device _names_.** Query capabilities at runtime (qubit count, native gates, connectivity) so your code survives the next hardware refresh.\n\n> `Try it` — Toggle \"Show retired devices\" and just read the descriptions. You're looking at the last five years of the quantum industry in one scrollable list.\n\n### Inside a device: the detail page\n\nClick any device name to open its detail page. The layout is consistent across the fleet, so learn it once. Up top is a **Summary** card:\n\n- **Status**, **Device ARN** (the string your code needs. Copy it here), and **Spending limit**.\n- Three **queue depth** counters: _Tasks_, _Priority tasks_, and _Hybrid jobs_. These tell you how many workloads are ahead of yours right now.\n- Action buttons: **Reserve device** (jump the queue via Braket Direct), **Open notebook** (spin up a notebook pre-pointed at this device), and a refresh icon.\n\nBelow that, up to four tabs. Let's use two real devices to see the range.\n\n**QuEra Aquila — a 256-qubit analog machine.** Aquila is a neutral-atom device that operates in _analog_ mode (continuous evolution of a Hamiltonian rather than discrete gates). Its detail page shows the hardware provider (QuEra, Boston), the Region (`us-east-1`), and, importantly, its cost and availability:\n\n![The detail page for QuEra Aquila, a 256-qubit analog neutral-atom QPU in us-east-1: online with 5 tasks queued, priced at $0.30 a task plus $0.01 a shot or $2,500 an hour reserved, with its weekly availability windows.](https://quantumdev.io/blog/media/image-212156c7b7638757f654dc38876121cc79bf02cc-2348x1648-webp-960.webp)\n\nThe **Calibration** tab is where hardware honesty lives. For Aquila you get device geometry and error characteristics — 256 sites, a 75 µm × 76 µm lattice, 4 µm spacing, position resolution down to 0.01 µm, plus filling, vacancy, and detection error rates — with a **Device geometry / Control fields / JSON** switch so you can grab the raw numbers programmatically. Every calibration table is timestamped (\"Last updated…\") because these values drift and are re-measured regularly.\n\n![Aquila's Calibration tab, a timestamped table of device geometry and error rates: 256 sites, a 75 by 76 micrometre lattice, and filling, vacancy, atom loss and detection errors.](https://quantumdev.io/blog/media/image-8b697e4b93535206f25cfe8fcf1595f4dfa3c0fe-1692x1504-webp-960.webp)\n\nThe **Experimental capabilities** tab exposes cutting-edge features you enable per-task through the SDK. For Aquila, _local detuning_, _tall geometries_, and _tight geometries_ (ways to push past the standard operating modes).\n\n![Aquila's Experimental capabilities tab, offering local detuning, tall geometries and tight geometries.](https://quantumdev.io/blog/media/image-4844b75e2002f772502f9666a0d1b4213383a5ce-2854x1250-webp-960.webp)\n\n**AQT IBEX Q1 — a 12-qubit gate-model machine.** On a gate-based QPU, the **Native Gates** tab lights up. It lists the exact gate set the hardware executes natively. Everything else in your circuit gets compiled down to these. For AQT's ion trap that's `prx` (a phase-RX gate), `xx` (an Ising XX coupling gate), and `rz` (a Z-axis rotation). Notice this device lives in `eu-north-1` This is a reminder that **QPUs are Region-specific.**\n\n![The Native Gates tab for AQT IBEX Q1, a 12-qubit ion-trap QPU in eu-north-1, listing prx, xx and rz, under a banner saying no spending limit is set.](https://quantumdev.io/blog/media/image-7f836a09aa560d0d0b653f21ad114e70ddaa2ec4-2868x1422-webp-960.webp)\n\nHere's how you'd pull all of that from code instead of the console:\n\n```python\nfrom braket.aws import AwsDevice\n\n# Copy the ARN straight from the device page's Summary card\naquila = AwsDevice(\"arn:aws:braket:us-east-1::device/qpu/quera/Aquila\")\n\nprint(aquila.status)                       # ONLINE / OFFLINE\nprint(aquila.properties.paradigm)          # qubit count, geometry, etc.\n# Gate-model devices expose their native gate set:\ngarnet = AwsDevice(\"arn:aws:braket:eu-north-1::device/qpu/iqm/Garnet\")\nprint(garnet.properties.paradigm.nativeGateSet)\n```\n\n| Tab | Appears on | What it tells you |\n| --- | --- | --- |\n| **Details** | All devices | Provider, region, qubits, cost, availability windows |\n| **Calibration** | All devices | Live, timestamped error rates and topology |\n| **Native Gates** | Gate-model QPUs only | The exact gate set your circuit compiles to |\n| **Experimental capabilities** | Select devices | Per-task opt-in features (dynamic circuits, geometry tricks) |\n\n> `Try it` — Open SV1 (free) and a QPU like IQM Garnet side by side. Compare their Calibration tabs. The simulator is \"perfect\"; the QPU is informatively imperfect.\n\n## 4. Notebooks: your Managed Jupyter Cockpit\n\nYou _can_ run Braket from your laptop with `pip install amazon-braket-sdk`. But another option is a **managed notebook** like a JupyterLab instance AWS spins up with the Braket SDK and the major quantum frameworks already installed.\n\nClick **Notebooks** and Braket offers two paths.\n\n![The Notebooks page, offering a standard notebook with a Create button and an advanced notebook with a Configure button.](https://quantumdev.io/blog/media/image-9aafe0fe27255170f9f9cdeb2b51ecc6670f51e9-3434x1276-webp-960.webp)\n\n|  | **Standard notebook** | **Advanced notebook** |\n| --- | --- | --- |\n| Setup | One click | 5-step wizard |\n| Compute | `ml.t3.medium` | EC2 / HPC / GPU, your choice |\n| Pre-installed | Braket SDK, CUDA-Q, PennyLane, Qiskit | Everything in Standard, plus… |\n| Extras | JupyterLab extensions, example algorithms | Custom IAM, VPC, KMS encryption, custom GitHub repos |\n| Best for | Learning, prototyping | Team environments, private networking, big compute |\n\n### The one-click standard notebook\n\nChoosing **Standard → Create** pops a small modal: name your instance (Braket prefixes it with `amazon-braket-`, max 35 characters, hyphens allowed, no spaces) and choose whether to enable the **Notebook inactivity manager.**  This is a helpful watchdog that stops the instance after idle time so you stop paying for compute you're not using.\n\n![The Create a notebook dialog: a name after the fixed prefix amazon-braket-, and the notebook inactivity manager set to enable or disable.](https://quantumdev.io/blog/media/image-5c435fcb05bf599816a3a095db35096706f1ff08-1614x646-webp-960.webp)\n\n> **Cost reality check:** A notebook instance is billed like the underlying EC2 instance for as long as it's _running_, whether or not you're using it. Enable the inactivity manager, or make a habit of **Stop**-ping the instance when you walk away. This is the most common source of a surprise Braket bill and it's not the quantum part.\n\n### The advanced notebook wizard, step by step\n\nClick **Advanced → Configure** for the full five-step wizard. This is where teams tailor an environment.\n\n![The five steps of the advanced notebook wizard: 1, general settings (name, theme, idle); 2, notebook configuration (instance type, volume, repository); 3, access permissions (IAM, root, KMS); 4, network access (VPC); 5, review and launch.](https://quantumdev.io/blog/media/image-997c821f19d2d8084c05df2ea2df10587bf9ff5e-1120x160-webp-960.webp)\n\n**Step 1 — General settings.** Name, light/dark visual mode, and the same inactivity manager toggle.\n\n![Step 1 of the notebook wizard, General settings: the notebook name, a light or dark visual mode and the inactivity manager.](https://quantumdev.io/blog/media/image-3c8a88bfe8fed7e2875e9752ee23ca51c2f4aa0e-2852x1120-webp-960.webp)\n\n**Step 2 — Notebook configuration.** Pick your compute. This is the step that decides your hourly cost and how much classical horsepower sits next to your quantum code.\n\nYou also set a **volume size** (5 GB up to 16 TB) and can **import a GitHub repository** so your notebook starts with your team's code already cloned.\n\n![Step 2 of the notebook wizard, Notebook configuration: Standard (ml.t3.medium), High performance (ml.c5.xlarge), GPU accelerated (ml.g4dn.xlarge) or a custom instance type, a 5 GB volume and an optional GitHub repository.](https://quantumdev.io/blog/media/image-2adac3eb2ea19dd3abc27e6e88bc005d1efedff2-2862x1360-webp-960.webp)\n\n**Step 3 — Access permissions.** Choose the IAM role the notebook assumes (e.g., the default `AmazonBraketServiceSageMakerNotebookRole`), decide on **root access**, and optionally supply a **KMS key** to encrypt the volume.\n\n![Step 3 of the notebook wizard, Access permissions: the existing role AmazonBraketServiceSageMakerNotebookRole, root access enabled and no custom encryption key.](https://quantumdev.io/blog/media/image-23609b70c6cb3535fbcc941c0aca1ab330667cc4-2864x1442-webp-960.webp)\n\n**Step 4 — Network access.** Attach the notebook to a **VPC** (or \"No VPC\" for a public instance). This is how you keep a notebook on a private network with no internet exposure.\n\n![Step 4 of the notebook wizard, Network access, with the VPC set to No VPC.](https://quantumdev.io/blog/media/image-0390b3c22df1112b578100fb4c762a0b01d47d54-2868x682-webp-960.webp)\n\n**Step 5 — Review and launch.** A tidy summary of every choice, plus optional resource **Tags** (great for cost allocation). Hit **Launch** and Braket provisions the instance.\n\n![Step 5 of the notebook wizard, Review and launch: the notebook amazon-braket-demo on ml.t3.medium with a 5 GB volume, root access and no VPC, with optional tags and a Launch button.](https://quantumdev.io/blog/media/image-78ba215cd48e0c4458f3a8f8578b8ba5a76b6664-2860x1622-webp-960.webp)\n\n### Your first circuit (once the notebook opens)\n\nOpen the launched notebook, start a `conda_braket` kernel, and here's the genuine \"hello, world\" of quantum computing: a **Bell state** on the free local simulator:\n\n```python\nfrom braket.devices import LocalSimulator\nfrom braket.circuits import Circuit\n\n# Build: an equal superposition of |00> and |11>\nbell = Circuit().h(0).cnot(0, 1)\nprint(bell)\n\n# Test: run on the free local simulator (up to ~25 qubits)\nresult = LocalSimulator().run(bell, shots=1000).result()\nprint(result.measurement_counts)      # ~ Counter({'00': 503, '11': 497})\n```\n\nTwo entangled qubits, measured a thousand times, coming out as roughly half `00` and half `11` — and never `01` or `10`. That's entanglement, and it cost you nothing. When you're ready to leave the simulator, you change exactly one line: the device. THIS is the whole point of Braket's design. We'll do that in [§6](#6-quantum-tasks-the-atomic-unit-of-work).\n\n## 5. Hybrid Jobs: Quantum-Classical, Orchestrated\n\nMost useful near-term quantum algorithms (QAOA, VQE, quantum machine learning) are _hybrid_: a classical optimizer runs a loop, and inside each iteration it calls a QPU. **Hybrid Jobs** is Braket's managed orchestration for exactly that pattern. It provisions a classical instance to run your driver code, gives your job **priority access** to the chosen QPU (so you're not re-queuing every iteration), streams metrics, checkpoints progress, and tears everything down when you're done.\n\nThe **Hybrid jobs** page starts empty, with **Create hybrid job** front and center. Creating one is a five-step wizard.\n\n![The empty Hybrid Jobs page, with a Create hybrid job button.](https://quantumdev.io/blog/media/image-ffa9c58a206f697553abb473e655aa3edfd9289e-3430x1116-webp-960.webp)\n\n**Step 1 — Specify an algorithm.** Upload a Python script from your computer or point at one in S3, set the **entry point** (the function invokes when the job starts), and optionally mark it **Zipped**. You can also expand **Hyperparameters** (values passed into your algorithm) and **Input data** (extra files staged for the job).\n\n![Step 1 of creating a hybrid job, Specify an algorithm: a zipped Python file uploaded from the computer, its entry point, and optional hyperparameters and input data.](https://quantumdev.io/blog/media/image-cf0ee5ea08150c4fb26a8aa713a6c1928644ca98-3414x1718-webp-960.webp)\n\n**Step 2 — Set hybrid job settings.** Name the job (defaults to `braket-job-default`) and, under **Advanced settings**, choose the execution role and the **default S3 bucket** where Braket stashes your code, inputs, outputs, and checkpoints.\n\n![Step 2 of creating a hybrid job: the job name braket-demo, the AmazonBraketJobsExecutionRole role and the default S3 bucket.](https://quantumdev.io/blog/media/image-7c65e57b6659310eefbb89c2ed7c75db26536d89-3430x1264-webp-960.webp)\n\n**Step 3 — Configure an environment.** Pick the container your code runs in. The base images ship with the Braket SDK, Python, and PennyLane; the TensorFlow and PyTorch variants add those frameworks (and CUDA); CUDA-Q adds GPU-accelerated simulation; or bring your own container (BYOC).\n\n![Step 3 of creating a hybrid job, Configure an environment: PennyLane, PennyLane with TensorFlow, PennyLane with PyTorch, CUDA-Q, or a custom container.](https://quantumdev.io/blog/media/image-d3538f3cc648ea185a8e3fe462e3c909126074b3-3436x1374-webp-960.webp)\n\n| Environment | Adds | Reach for it when… |\n| --- | --- | --- |\n| **PennyLane** | Core Braket SDK + PennyLane | Standard variational algorithms |\n| **PennyLane + TensorFlow** | TF + CUDA | ML/AI-flavored quantum work |\n| **PennyLane + PyTorch** | PyTorch | Computer-vision / NLP-style pipelines |\n| **CUDA-Q** | NVIDIA CUDA-Q | GPU-accelerated quantum simulation |\n| **Custom (BYOC)** | Your Docker image | You need exact, reproducible deps |\n\n**Step 4 — Configure the execution.** Under **Quantum compute settings** you pick the **Device** your job gets priority on (or use an _embedded simulator_ inside the classical instance), and optionally attach a **Reservation** to bypass queues entirely. Under **Classical compute settings** you size the driver instance. Then set a **Stopping condition** (max runtime default to 5 days) and optionally configure **Checkpoints** and **Outputs**.\n\n![Step 4 of creating a hybrid job, Configure the execution: the quantum device and an optional reservation, a Standard ml.m5.large instance, 5 GB of extra storage, a maximum runtime of 5 days, and optional checkpoints and outputs.](https://quantumdev.io/blog/media/image-0119078b0bfb2718c19058ff941a38d4589c7d21-2934x1704-webp-960.webp)\n\n| Classical instance | Specs | Use |\n| --- | --- | --- |\n| Standard | `ml.m5.large` — 2 vCPU, 8 GiB | Most driver loops |\n| GPU accelerated | `ml.g4dn.xlarge` — 4 vCPU, 16 GiB, 1× NVIDIA T4 | Embedded GPU simulators, ML |\n| Advanced selection | any supported type | Full control |\n\n**Step 5 — Review and create.** Everything in one screen from algorithm source, and role, to environment, hardware, and device ARN. Notice you can build and validate an entire hybrid job pointed at the **local simulator** (`local:braket/default-simulator`) so you never spend a cent debugging orchestration before you point it at real hardware.\n\n![Step 5 of creating a hybrid job, Review and create: the job braket-demo on ml.m5.large with the PennyLane runtime, pointed at the local simulator local:braket/default-simulator.](https://quantumdev.io/blog/media/image-66684fb0bee5aed62aafae473654f17f8ccc2a85-2650x1346-webp-960.webp)\n\nThe console wizard mirrors a decorator-based API. In a notebook, the same job is a few lines:\n\n```python\nfrom braket.jobs import hybrid_job\nfrom braket.devices import LocalSimulator\n\n@hybrid_job(device=\"local:braket/default-simulator\")\ndef run_job():\n    from braket.circuits import Circuit\n    circ = Circuit().h(0).cnot(0, 1)\n    return LocalSimulator().run(circ, shots=1000).result().measurement_counts\n\njob = run_job()          # Braket provisions, runs, and tears down\nprint(job.result())\n```\n\n> `Try it` — Create a hybrid job pointed at the local simulator with the tiny script above. It costs nothing and teaches you the entire lifecycle — provisioning, running, checkpointing, results in S3 — before real hardware is involved.\n\n## 6. Quantum Tasks: The Atomic Unit of Work\n\nA **quantum task** is the smallest thing Braket runs: one program (a circuit, an annealing problem, or an analog schedule) submitted to one device for a set number of **shots** (repetitions). Every simulator run, every QPU run, every iteration inside a hybrid job; all of it is quantum tasks under the hood. The **Quantum tasks** page is your run history.\n\nA banner at the top makes a point worth remembering (YET AGAIN): **QPUs are Region-specific.** A task for IQM Garnet must be submitted in `eu-north-1`; a task for QuEra Aquila in `us-east-1`. If your task list looks empty, check your Region first.\n\n![The Quantum Tasks page listing four completed tasks on IQM Garnet and IQM Emerald in eu-north-1, under a note that QPUs are Region-specific.](https://quantumdev.io/blog/media/image-27632db568eb6ba1a4ca268f502822f9bec821ce-3430x1084-webp-960.webp)\n\nClick a task to open its detail page. This is the receipt for a single run:\n\n- **Quantum task ARN** and **Device ARN** — the unique identifiers.\n- **Status** — `CREATED → QUEUED → RUNNING → COMPLETED` (or `FAILED`/`CANCELLED`).\n- **Successful shots** — e.g., `100 / 100 (100%)`.\n- **Queue position** — where you sit when queued.\n- **Results** — a **View in S3** link; raw measurement data always lands in your Braket S3 bucket.\n- **Reservation ARN** / **Hybrid job ARN** — populated if the task ran under either.\n- Tabs for a **Results explorer** and **Tags**.\n\n![A quantum task's detail page: a completed program on IQM Garnet with 100 of 100 shots successful, a View in S3 link to the results, and tabs for the Results explorer and Tags.](https://quantumdev.io/blog/media/image-4aab2237dba5157d5a4156fc1d226c76b6ee3753-3430x1524-webp-960.webp)\n\nNow the one-line change that turns your free Bell-state test into a real quantum-hardware run — swap `LocalSimulator()` for an `AwsDevice` ARN. This is the moment the guide has been building toward:\n\n```python\nfrom braket.aws import AwsDevice\nfrom braket.circuits import Circuit\nfrom braket.tracking import Tracker      # near-real-time cost tracking\n\nbell = Circuit().h(0).cnot(0, 1)\n\n# On-demand cloud simulator (SV1) — cheap, region-agnostic ARN:\nsv1 = AwsDevice(\"arn:aws:braket:::device/quantum-simulator/amazon/sv1\")\n\n# ...or a real QPU (must be in that device's Region!):\ngarnet = AwsDevice(\"arn:aws:braket:eu-north-1::device/qpu/iqm/Garnet\")\n\nwith Tracker() as tracker:\n    task = garnet.run(bell, shots=100)          # returns immediately\n    print(task.state())                          # QUEUED / RUNNING / COMPLETED\n    counts = task.result().measurement_counts    # blocks until done\n    print(counts)\n\nprint(tracker.quantum_tasks_statistics())        # shots, tasks, duration\nprint(tracker.qpu_tasks_cost())                  # estimated $ before the bill\n```\n\nThe `Tracker` context manager is worth its weight in gold. It gives you a **maximum cost estimate in USD in near-real-time**, so you know what a run cost before the monthly bill confirms it.\n\n> `Try it` — Run the Bell state on `SV1` first (cloud simulator, cents). When the counts look right, change one line to a QPU ARN and watch your task move through `QUEUED → RUNNING → COMPLETED` on the Quantum tasks page. Congratulations — you just used a real quantum computer.\n\n## 7. Algorithm Library: 11 Ready-to-Run Classics\n\nIf you learn better by reading working code, **Algorithm library** is the best page on the console. It's a catalog of pre-built, textbook quantum algorithms, each written in Python, each with a **GitHub** link, each runnable in a managed notebook via **Open notebook**. No blank-page problem. You start from a correct implementation and modify.\n\n![The Algorithm library's 11 algorithms, each with a GitHub link: Bernstein-Vazirani, Deutsch-Jozsa, Grover's, QAOA, the quantum circuit Born machine, the quantum Fourier transform, quantum phase estimation, quantum walk, Shor's, Simon's and a violation of Bell's inequality.](https://quantumdev.io/blog/media/image-6c27d6bcd9e40658695fb2349a05c2202ab177c5-1686x1646-webp-960.webp)\n\n| Algorithm | What it demonstrates |\n| --- | --- |\n| **Bernstein–Vazirani** | The first algorithm to beat classical for a specific problem — oracle separation |\n| **Deutsch–Jozsa** | An early, clean quantum-vs-classical advantage |\n| **Grover's** | Quadratic speedup for unstructured search (√N vs N) |\n| **QAOA** | Hybrid optimization — the workhorse of the NISQ era |\n| **Quantum Circuit Born Machine** | Variational generative modeling |\n| **Quantum Fourier Transform** | The subroutine behind Shor's and phase estimation |\n| **Quantum Phase Estimation** | Eigenvalue estimation — foundational to many algorithms |\n| **Quantum Walk** | The quantum analog of a random walk |\n| **Shor's** | Polynomial-time integer factoring — the famous one |\n| **Simon's** | Exponential separation; a precursor to Shor's |\n| **Violation of Bell's Inequality** | Proof that entanglement is real, in code |\n\n> `Try it` — Open **Grover's algorithm**, run it on the local simulator, then change the \"marked\" item and watch the probability concentrate on your new target. It's the most satisfying five minutes in the console.\n\n## 8. Braket Direct: Skip the Line and Go\n\n**Braket Direct** is a hub of three \"power-user\" programs bundled together: dedicated hardware access, human expertise, and bleeding-edge features.\n\n![The Braket Direct page: reservations; expert advice from Braket office hours, IonQ, QuEra, Rigetti and the Advanced Solutions Lab; and experimental capabilities on IQM Emerald, IQM Garnet and QuEra Aquila.](https://quantumdev.io/blog/media/image-2efcd7cdf9fa7da33e6ace5480e3197607766fbf-1686x1506-webp-960.webp)\n\n**Reservations** give you _exclusive_ access to a specific QPU for a block of time, so your workloads don't wait behind anyone else's. Key facts (verified against the Braket developer guide):\n\n- Available in **1-hour increments** for all Braket devices.\n- **Cancellable up to 48 hours in advance** at no charge; otherwise you pay for the reserved block regardless of how much you use it.\n- Queue tasks and jobs _in advance_ against your **Reservation ARN**, or submit live during the window.\n- IonQ perk: reservations raise the per-circuit gate limit to 5,000 (vs. 2,000 on-demand) and lower the error-mitigation minimum shot count to 500 (vs. 2,500).\n\nClicking **Reserve device** opens a request form — your name and email, a description of your workload, a duration (1–5+ hours), two or more candidate dates, and an option for a free 30-minute prep session with a Braket expert.\n\n![The Create reservation form: name, email, a description of the workload, a duration from 1 to 5 or more hours, two or more candidate dates, and an optional 30-minute prep session.](https://quantumdev.io/blog/media/image-95f4a6f319808e69944b73e179646f0819d6803d-1216x1954-webp-960.webp)\n\n**Expert advice** connects you to real humans: free weekly **Braket office hours** (30-minute 1:1 slots), plus dedicated channels to **IonQ, QuEra, and Rigetti**, and the **Advanced Solutions Lab (ASL)** for deeper collaborative research engagements.\n\n**Experimental capabilities** surfaces next-generation features with limited availability — enabled per-task through the SDK. In this snapshot: **dynamic circuits** (mid-circuit measurement and feed-forward) on **IQM Emerald** and **IQM Garnet**, and **local detuning / tall geometries / tight geometries** on **QuEra Aquila**.\n\n| Program | What you get | Cost model |\n| --- | --- | --- |\n| **Reservations** | Exclusive device access, no queue | Per-hour, billed for the block |\n| **Expert advice** | Office hours, provider channels, ASL | Office hours free; ASL varies |\n| **Experimental capabilities** | Early hardware features | Per-task, standard device pricing |\n\n> `Try it` — Even if you never reserve hardware, book a **Braket office hours** slot. Thirty free minutes with an AWS quantum specialist is the highest-leverage button on this entire page.\n\n## 9. Program Sets: Batch many circuits into one task\n\nA mid-2026 addition worth calling out: **Program Sets.** Historically each circuit was its own quantum task with its own per-task fee and its own trip through the queue. A program set lets you bundle **up to 100 unique circuits** _or_ a single **parametric circuit with up to 100 parameter settings** into **one** quantum task. Braket handles compiling and running the whole package on the QPU, minimizing the gap between executions and the per-task overhead.\n\nWhy you care:\n\n- **One per-task fee** across every circuit in the set (you still pay per shot).\n- **Less queue waiting** — one submission instead of a hundred.\n- Ideal for **variational algorithms, classical shadows, and quantum machine learning**, where you sweep a parameter or run a batch of related circuits.\n\nProgram sets are available across the gate-based devices. If your workload is \"run this circuit 100 times with different angles,\" this is the feature that makes it cheap and fast.\n\n## 10. Cost Controls\n\nQuantum hardware is metered, and QPU time is not cheap. Braket gives you three independent guardrails. Use all of them.\n\n### How does Braket charge?\n\nYou pay for what you run:\n\n| Resource | How it's billed |\n| --- | --- |\n| **Local simulator** (in the SDK) | **Free** |\n| **On-demand simulators** (SV1, DM1) | Per **minute** of simulation time |\n| **QPUs** | Per **task** + per **shot** (rates vary by device) |\n| **Reservations** | Per **hour** of the reserved block, regardless of use |\n| **Notebooks / hybrid job classical compute** | Per the underlying EC2 instance, while running |\n\nA concrete QPU example straight off the device page: QuEra Aquila: **$0.30 per task + $0.01 per shot** on-demand, or **$2,500 per hour** reserved. So a 100-shot Aquila task runs about `$0.30 + 100 × $0.01 = $1.30`. Rates differ per device and change over time, so always confirm current numbers on the [Amazon Braket pricing page](https://aws.amazon.com/braket/pricing/) or the device's Details tab before a big run.\n\n### Guardrail 1 — Spending limits (hard caps)\n\nUnder **Administration → Spending limits** you set a maximum dollar amount per QPU. When spend hits the cap, further tasks to that device are **rejected**. Because QPUs are Region-specific, so are spending limits: select the device's Region to see and set them.\n\n![The Spending limits page with one limit: IQM Garnet, capped at $150.00, with $1.34 spent.](https://quantumdev.io/blog/media/image-cd1437054d84b5bc38bd18df161e11482f3e96ba-3432x1170-webp-960.webp)\n\n**Create spending limit** opens a small form.\n\n![The Create spending limit dialog: a device, a maximum amount in US dollars and an optional time period.](https://quantumdev.io/blog/media/image-27ca4e89332ac53f758f956ae21ff70133ff9271-1164x762-webp-960.webp)\n\n### Guardrail 2 — In-code cost tracking\n\nThe SDK's `Tracker` (shown in [§6](#6-quantum-tasks-the-atomic-unit-of-work)) gives you a **near-real-time maximum cost estimate in USD** while your code runs. Wrap any run in `with Tracker() as tracker:` and call `tracker.qpu_tasks_cost()` / `tracker.simulator_tasks_cost()`. Every Braket example notebook already includes it.\n\n### Guardrail 3 — Operational hygiene\n\n- **Stop idle notebooks** (or enable the inactivity manager) — this is the most common surprise charge, and it's classical compute, not quantum.\n- **Set a** `Maximum runtime` **stopping condition** on every hybrid job so a runaway optimizer can't loop forever.\n- **Test on simulators first.** Local is free; on-demand SV1/DM1 are cheap. Only graduate to a QPU when the circuit is correct.\n- **Tag** notebooks, tasks, and jobs so AWS Cost Explorer can slice spend by project.\n\n> `Try it` — Before your very first QPU submission, set a $10 spending limit on that device. You'll run your experiment and physically cannot overspend. Raise it later when you trust your workflow.\n\n## 11. Announcements: how to stay current\n\nBecause the fleet changes monthly, **Announcements** are very helpful. It's the changelog you actually need. It's where AWS posts new devices, retirements (with hard cutoff dates), and feature launches.\n\n![The Announcements page: program sets on IonQ devices, the retirement of the TN1 simulator on 26 June 2026, and longer availability hours for AQT IBEX Q1.](https://quantumdev.io/blog/media/image-43b3a9b7779868bb376a3fcc377f9b42bc647a53-3428x1692-webp-960.webp)\n\n> `Try it` — Skim Announcements once a month, or follow the AWS Quantum Computing Blog (linked from the dashboard's Community support widget). A retirement notice you missed is a broken pipeline you didn't see coming.\n\n## 12. Your first hour on Braket: a hands-on path\n\nEnough touring. Here's a concrete sequence that takes you from zero to \"I ran on a real quantum computer,\" safely.\n\n1. **Enable devices (2 min).** Administration → Permissions and settings → enable third-party devices; verify the three execution roles.\n2. **Set a guardrail (1 min).** Administration → Spending limits → create a small cap (say $10) on one QPU you want to try.\n3. **Browse the fleet (5 min).** Devices → open a simulator and a QPU; compare their Calibration tabs; copy a device ARN.\n4. **Launch a notebook (5 min).** Notebooks → Standard → Create; enable the inactivity manager; open JupyterLab with the `conda_braket` kernel.\n5. **Build + test for free (10 min).** Run the Bell-state circuit on `LocalSimulator()`, then on cloud `SV1`. Confirm the counts.\n6. **Run on real hardware (10 min).** Change one line to a QPU ARN (mind the Region!), wrap it in `Tracker()`, submit, and watch the task move `QUEUED → RUNNING → COMPLETED`.\n7. **Analyze (5 min).** Quantum tasks → open your task → **View in S3** for raw results; check `tracker.qpu_tasks_cost()`.\n8. **Learn from the pros (rest of the hour).** Algorithm library → open Grover's or QAOA and modify it. Book a Braket office-hours slot.\n\n## 13. Wrap-up: go press the buttons\n\nThe thing that surprises most developers about Amazon Braket is how straightforward the experience is. It's the same console muscle memory as any other AWS service: a left-nav, some wizards, IAM roles, an S3 bucket, a spending limit. The only difference is that the \"compute\" at the end of the pipe happens to be trapped ions in a vacuum chamber or neutral atoms held by lasers. You build and test for free, you flip one line to reach real hardware, and you can cap the cost so tightly that experimentation is genuinely low-risk.\n\nSo don't just read about it. Open the console, enable devices, set a $10 limit, and run a Bell state on a real QPU this week. The gap between \"quantum computing is fascinating\" and \"I ran a quantum program\" is about a few minutes and a handful of clicks.\n\nNow you have the map.\n\nLet's keep building.\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3GaxVTZeaL9pWzjXj3k7tMynzbI/a-developers-field-guide-to-amazon-braket) on 17 July 2026.\n"
  },
  {
   "id": "post:unlocking-superposition-the-hadamard-gate-on-amazon-braket",
   "type": "post",
   "title": "Unlocking Superposition: The Hadamard Gate on Amazon Braket",
   "url": "https://quantumdev.io/blog/unlocking-superposition-the-hadamard-gate-on-amazon-braket/",
   "markdownUrl": "https://quantumdev.io/blog/unlocking-superposition-the-hadamard-gate-on-amazon-braket.md",
   "description": "Learn to build your first quantum circuit on AWS. Master the Hadamard gate, explore superposition, and get started with Amazon Braket.",
   "published": "2026-06-28",
   "updated": "2026-06-28",
   "author": "Christian Perez",
   "text": "# Unlocking Superposition: The Hadamard Gate on Amazon Braket\n\nBy Christian Perez, published 2026-06-28\n\nCanonical: https://quantumdev.io/blog/unlocking-superposition-the-hadamard-gate-on-amazon-braket/\n\nLearn to build your first quantum circuit on AWS. Master the Hadamard gate, explore superposition, and get started with Amazon Braket.\n\nYou have probably spent plenty of time with classical computing on AWS. It's likely you already understand logic gates like AND, OR, and NOT. These are the simple binary operations that form every program ever written.\n\n[Quantum gates](https://quantum.altivum.ai/glossary/quantum-gate) play the exact same role in quantum computing. However, they operate under fundamentally different rules. Understanding them is an absolute requirement. If you want to build anything real on [Amazon Braket](https://aws.amazon.com/braket/), quantum gates are the place where the actual computation happens.\n\nSome of the vocabulary I introduce in this article can be a bit complex or foreign but I have linked as many references as possible in order to help you, the reader and builder, begin to understand some of these critical concepts in quantum computing.\n\n## What is a Quantum Gate?\n\nA quantum gate is a single operation that manipulates one or more [qubits](https://quantum.altivum.ai/glossary/qubit). To truly grasp this concept, we need to first look at the qubit itself.\n\nAs I've mentioned in earlier articles in this series, a qubit is the quantum equivalent of a classical bit. Classical bits are permanently locked to a state of either 0 or 1. A qubit, however, can exist in a superposition or both states simultaneously. This dual state is represented mathematically as a probability amplitude stretching across the entire state space.\n\nMathematically, quantum gates are unitary matrices. For a gate U to be valid, it must satisfy U†U = \"I\", where U† is what is called a conjugate transpose (more on this below). \"I\" is the identity matrix. This requirement ensures that total probability is always conserved. In other words, you can never \"lose\" a quantum state by passing it through a gate. What this implies is that all quantum gates are reversible, which is very different when compared to classical gates like AND or OR.\n\nI know that's pretty dense so let's break it down into two simpler concepts:\n\n- [Unitary Matrix](https://quantum.altivum.ai/glossary/unitary-matrix): A square matrix that doesn’t change the length of vectors when it acts on them. It’s a \"length‑preserving\" linear transformation.\n- Conjugate Transpose: This is simpler than it sounds. First, you _transpose_ the matrix (turn rows into columns). Then, you take the _complex conjugate_: for every entry that looks like a+bi, you change the sign on the imaginary part to get a-bi.\n\nSo again, a quantum gate has to fit the criteria of a unitary matrix.\n\nWhen you chain gates together in sequence, you're constructing a quantum circuit: a series of reversible operations that transforms an initial state into a measurable result.\n\n## What is the Bloch sphere?\n\nBefore we jump into specific gates, it helps to have a mental model of what a qubit actually looks like when it changes states. This is where the [Bloch Sphere](https://quantum.altivum.ai/glossary/bloch-sphere) comes in.\n\n![A geometric representation of a single qubit's probability space.](https://quantumdev.io/blog/media/image-7c1735ed9f968afee4415a60dd9780ff0f1179a0-2048x2048-webp-960.webp)\n_Image by Higgsfield.ai_\n\nThe Bloch sphere is a geometric representation of a single qubit's probability space. Think of it like a globe. At the very top (the North Pole), you have the state |0⟩ (pronounced \"ket zero\"), which is the ground state or the equivalent of a classical 0. At the very bottom (the South Pole), you have the state |1⟩ (pronounced \"ket one\"), the equivalent of a classical 1.\n\nUnlike a classical bit that can only exist at one of those two poles, a qubit can exist anywhere on the surface of this sphere. Any point on the sphere represents a valid, unique quantum state (a specific superposition of |0⟩ and |1⟩).\n\nSo why is this important?\n\nThe Bloch sphere is crucial because it takes the math of quantum mechanics and turns it into an intuitive geometry problem. When building circuits, instead of trying to visualize unitary matrix multiplication in your head, you can think of quantum gates as physical rotations.\n\nWhen a gate acts on a qubit, it is simply grabbing that point on the globe and rotating it to a new location. For example, a quantum NOT gate is just a 180-degree flip that moves the state from the North Pole to the South Pole.\n\nHaving this mental model makes it much easier to grasp what your code is actually doing to the quantum hardware.\n\n## What does the Hadamard gate do?\n\nTo get started, we are only going to focus on one single operation: [The Hadamard Gate](https://quantum.altivum.ai/glossary/hadamard-gate).\n\nThe Hadamard gate is the gateway to superposition. When you apply it to a qubit in the ground state |0⟩ (pronounced “ket zero” also the North Pole of our Bloch sphere), it physically rotates that state down to the equator of the sphere.\n\n![Diagram of the Hadamard gate: the H symbol, then two Bloch spheres showing the state |0⟩ at the north pole turned by H into (|0⟩ + |1⟩)/√2 on the equator.](https://quantumdev.io/blog/media/image-033ef591e03ac9fda0dfc5065060d27480cd70a2-2048x2048-webp-960.webp)\n_H Gates \\& the Bloch Sphere_\n\nWhat happens at the equator? The qubit enters an equal superposition of |0⟩ and |1⟩ (pronounced “ket one”).\n\nSimply put, this means the qubit isn't just a 0 or a 1 anymore; it is existing in a fluid state where it has an exactly 50% probability of being measured as a 0, and a 50% probability of being measured as a 1.\n\nThis is step zero for almost every quantum algorithm ever written. Classical computers have to evaluate possibilities one by one, like trying every key on a keyring until a door unlocks. But if you have a register of qubits and hit them with Hadamard gates, you create a massive computational landscape where every possible combination of 0s and 1s exists simultaneously.\n\nBefore a quantum computer can perform its magic and explore multiple solutions at once, the qubits have to be unlocked and put into this state of superposition first. The Hadamard gate is the spark that makes that happen.\n\n## From Physics to Code: Why the Hadamard Gate Unlocks Amazon Braket\n\nIt is easy to look at the Bloch sphere or the math of a Hadamard gate and think, “This is neat physics, but how does it help me build an application on AWS?”\n\nIn classical cloud computing, AWS gives you compute primitives like EC2 instances or Lambda functions. In quantum cloud computing, Amazon Braket gives you access to physical [Quantum Processing Units (QPUs)](https://quantum.altivum.ai/glossary/qpu) from providers like IonQ, Rigetti, and QuEra. But unlike a classical CPU, you cannot just send a Python script or a SQL query to a quantum computer and expect it to run.\n\nQPUs only understand physical manipulations like laser pulses or microwave bursts that change the states of atoms or superconducting loops.\n\nIn the cloud, we represent these physical manipulations as quantum circuits, and the foundational building block of those circuits is the Hadamard gate.\n\n## How do you write your first circuit on Amazon Braket?\n\nAmazon Braket is AWS's fully managed quantum computing service. [Quantum circuits ](https://quantum.altivum.ai/glossary/quantum-circuit)flow through the Braket Python SDK, where gates are first-class objects.\n\nLet's write a tiny circuit.\n\nWe will take a single qubit, apply a Hadamard gate to put it into superposition, and then look at the circuit.\n\n```python\nfrom braket.circuits import Circuit\n\n# Initialize a circuit and apply the Hadamard (H) gate to qubit 0\ncircuit = Circuit().h(0)\n\n# Print the visual representation of the circuit\nprint(circuit)\n```\n\nWithout gates, qubits do nothing. The circuit, which is the sequence of gates, is the algorithm itself.\n\n## How do you run your circuit?\n\nThe fastest way to start testing your quantum circuits is the [Amazon Braket Local Simulator](https://quantum.altivum.ai/glossary/localsimulator), which is free and runs entirely on your local machine.\n\nBecause a qubit in equal superposition has a 50% chance of being measured as a 0 and a 50% chance of being measured as a 1, we can't just run the circuit once to see how it works. We need to run it multiple times to build up a statistical distribution of results.\n\nRunning this circuit is called shots. So when you run 1000 shots of a circuit, you run that circuit 1000 times.\n\nHere is how you run your Hadamard circuit 1,000 times:\n\n```python\nfrom braket.devices import LocalSimulator\nfrom braket.circuits import Circuit\n\n# 1. Define the local simulator device\ndevice = LocalSimulator()\n\n# 2. Build the circuit: Apply Hadamard to qubit 0\ncircuit = Circuit().h(0)\n\n# 3. Run the circuit for 1000 shots\ntask = device.run(circuit, shots=1000)\n\n# 4. Fetch and print the results\nresults = task.result()\nprint(results.measurement_counts)\n```\n\nWhat to expect: When you print the `measurement_counts`, you should see a result that looks something like `{'0': 502, '1': 498}`. Because quantum mechanics is probabilistic, you won't get exactly 500/500 (or 50%), but it will be very close.\n\nFrom here, the path to real Quantum Processing Units (QPUs) is the exact same code. You just swap `LocalSimulator()` for `AwsDevice()` with a specific QPU ARN, point your results to an S3 bucket, and you're suddenly running on actual quantum hardware in the cloud.\n\n## Key Takeaways\n\n- **Quantum gates are reversible transformations:** Unlike classical logic gates that destroy information, quantum gates are unitary operations that simply rotate a qubit's probability state without ever erasing data.\n- **The Bloch sphere is your visual map:** It translates complex quantum math into an intuitive 3D globe, where the North Pole is 0, the South Pole is 1, and the surface represents every possible state of superposition.\n- **The Hadamard gate is step zero:** This foundational gate moves a resting qubit to the equator of the Bloch sphere, creating a 50/50 superposition that unlocks a quantum computer's ability to explore multiple possibilities at once.\n- **Circuits are the language of the cloud:** Amazon Braket QPUs do not run classical scripts; they require quantum circuits (built from gates like the Hadamard) to physically manipulate quantum hardware.\n- **Build locally, execute globally:** The Amazon Braket SDK allows you to construct and test your circuits for free using a local simulator, then use that exact same code to run workloads on real quantum computers in the AWS cloud.\n\n_For more on Amazon Braket circuit construction, check out the official SDK documentation and the_ `amazon-braket-examples` _GitHub repository._\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3Fhr1XSi6oCtLlTq1RPLQCI6I45/unlocking-superposition-the-hadamard-gate-on-amazon-braket) on 28 June 2026.\n"
  },
  {
   "id": "post:fault-tolerance-quantums-hardest-problem-has-an-end-date",
   "type": "post",
   "title": "Fault Tolerance: Quantum's Hardest Problem Has an End-Date",
   "url": "https://quantumdev.io/blog/fault-tolerance-quantums-hardest-problem-has-an-end-date/",
   "markdownUrl": "https://quantumdev.io/blog/fault-tolerance-quantums-hardest-problem-has-an-end-date.md",
   "description": "Quantum's real bottleneck isn't qubit count. It's fault tolerance. How error correction, AWS Ocelot, and QuEra's Libra are ending the noisy era by 2028.",
   "published": "2026-06-20",
   "updated": "2026-06-20",
   "author": "Christian Perez",
   "text": "# Fault Tolerance: Quantum's Hardest Problem Has an End-Date\n\nBy Christian Perez, published 2026-06-20\n\nCanonical: https://quantumdev.io/blog/fault-tolerance-quantums-hardest-problem-has-an-end-date/\n\nQuantum's real bottleneck isn't qubit count. It's fault tolerance. How error correction, AWS Ocelot, and QuEra's Libra are ending the noisy era by 2028.\n\nAs we continue our journey into the world of quantum computing, we start to wonder why we haven't been able to leverage this technology at scale before, and why is this becoming a possibility now?\n\nMost quantum headlines you read talk about number of qubits. The number of qubits, however, does not address the real rate-limiting factor. What makes a quantum computer effective is how many operations a machine can run before its answer dissolves into noise. That limit is set by fault tolerance, and it is the problem that research in quantum computing is continuously trying to solve.\n\nQubits are fragile. Anything from a stray vibration, a disruption from a magnetic field, or the heat of the hardware itself can corrupt the quantum state in the middle of a calculation. Fault tolerance is the ability to compute correctly anyway. It runs on quantum error correction, which spreads one reliable \"logical\" qubit across many noisy physical ones. It then detects and fixes errors faster than they pile up.\n\nWhy does this matter? If you run a circuit on today's hardware, errors accumulate within a few hundred operations. The useful algorithms, the ones that crack molecular simulation, materials discovery, and large-scale optimization, need millions to billions of reliable operations. Fault tolerance enables us to run as many computations as is necessary to solve real-world problems.\n\nWe are currently in (but quickly nearing the end of) the noisy era. The signal of progress has moved from physical qubit counts to logical qubit counts. A thousand noisy qubits you cannot error-correct are useless. Given this fact, the obstacle becomes overhead: conventional surface codes can demand on the order of a thousand physical qubits to protect a single logical one. Our goal is to drive that ratio down, prove stable operation below threshold, and unlock the possibilities of quantum computing.\n\nAWS is pushing on both sides of that frontier. In its own labs, the AWS Center for Quantum Computing built Ocelot, a chip that bakes error correction into the hardware using cat qubits. Those qubits intrinsically suppress bit-flip errors, which leaves only phase-flip errors for a simple repetition code to mop up. (More on bit-flip and phase-flip errors in the comments). AWS reports up to 90% fewer qubits spent on correction, fabricated with standard microelectronics processes so the design can scale.\n\nOn Braket, AWS is bringing fault-tolerant hardware to the cloud through its deepened partnership with QuEra. The target is Libra in 2028. Libra is a megaquop machine running roughly one million reliable logical operations across more than 256 error-corrected logical qubits at a 10⁻⁶ logical error rate. It's built on reconfigurable neutral atoms. The destination is bigger than any single device though. AWS wants quantum sitting in the stack as a standard compute modality, alongside CPUs, GPUs, and AI accelerators.\n\nSo here is the takeaway: fault tolerance now has an end-date on the calendar. Start thinking in logical qubits and prototype hybrid quantum-classical workflows on Braket so the moment error-corrected hardware lands, you already know how to use it.\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3FQ75jn0dyqWrBJBNeShgpsBBX4/fault-tolerance-quantums-hardest-problem-has-an-end-date) on 20 June 2026.\n"
  },
  {
   "id": "post:dear-builder-welcome-to-quantum-computing",
   "type": "post",
   "title": "Dear Builder, Welcome to Quantum Computing",
   "url": "https://quantumdev.io/blog/dear-builder-welcome-to-quantum-computing/",
   "markdownUrl": "https://quantumdev.io/blog/dear-builder-welcome-to-quantum-computing.md",
   "description": "An invitation to start exploring what is possible with quantum computing on Amazon Braket.",
   "published": "2026-06-17",
   "updated": "2026-06-17",
   "author": "Christian Perez",
   "text": "# Dear Builder, Welcome to Quantum Computing\n\nBy Christian Perez, published 2026-06-17\n\nCanonical: https://quantumdev.io/blog/dear-builder-welcome-to-quantum-computing/\n\nAn invitation to start exploring what is possible with quantum computing on Amazon Braket.\n\nMy last article discussed quantum computing in great detail.\n\nI revisited that article and what I realized is that what I should've started with is a welcome letter. A welcome letter that invites all of us as builders to start thinking about the problems quantum computing can solve. Despite the fact that this is a relatively new technology, it is becoming commercially available fast (though I'd be doing you a disservice if I withheld the fact that we've had quantum computers since the late 1990s).\n\nSo let's start over.\n\nDear Builders,\n\nMany of you have provisioned more infrastructure than you can probably remember. Everything from S3 buckets, to Lambda functions, and SageMaker endpoints. There is, however, a service in your console, right now, that can connect to a machine cooled to temperatures near absolute zero, or to a cloud of atoms held in place by lasers. It uses the same IAM roles, same S3 buckets, same bill, and it's managed entirely by AWS.\n\nI'm talking about Amazon Braket.\n\nThis is your on-ramp to quantum computing. Welcome.\n\nThe reality is that, in many ways, we're early. Today, these machines are noisy, comprised of a few hundred qubits at most, and nowhere near replacing the classical compute behind your current architecture. For nearly everything you run, a quantum computer would be slower and pricier, and knowing that is the correct engineering call. But newer chips have shown that real error correction is finally possible.\n\nSo why care now? Two simple reasons:\n\nFirst, the inevitable capabilities that quantum computing unlocks. A large enough quantum computer can break the encryption behind most of the internet. Adversaries can harvest data now, and decrypt later, recording your traffic today to crack it tomorrow. Frankly, anything that must stay secret for a decade is already exposed. The National Institute of Standards and Technology (NIST) finalized post-quantum standards in 2024, and AWS already ships them: KMS, ACM, and Secrets Manager support hybrid post-quantum key exchange today. What this means for us as builders is that we can start to inventory long-lived secrets and make sure TLS clients negotiate ML-KEM (Module‑Lattice‑Based Key‑Encapsulation Mechanism).\n\nAnd second, the competitive moat is starting to form. Every big technology has an awkward phase where it works in a lab but not yet in production. The engineers who build fluency during that phase are early, and early compounds over time. There are maybe 30,000 people on Earth who can work in quantum today, against a projected 250,000 roles by 2030. The scarce skill isn't a PhD in Physics but rather the hybrid practitioner who can put together a quantum routine in a cloud environment. Someone who can bridge the gap between superpositio and IAM policies. Someone like you.\n\nStarting has never been cheaper. Braket is a managed service that leverages multiple types of machines: trapped-ion, superconducting, and neutral-atom hardware all from several vendors, behind one SDK and one bill. You debug on free local simulators, then graduate to real hardware for a few dollars a run. The architecture is hybrid composed of a classical optimizer looping with a quantum backend, orchestrated by Braket Hybrid Jobs, and results landing in your S3 bucket.\n\nI encourage you to check it out. Open Braket, launch a notebook, and run a Bell state on the free simulator.\n\nWelcome to the era of AI and Quantum Computing.\n\nSincerely,\n\nA fellow builder.\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3FFVnDe3HIsgHRwAz0zPsKBgjZv/dear-builder-welcome-to-quantum-computing) on 17 June 2026.\n"
  },
  {
   "id": "post:the-state-of-the-tech-quantum-computing-on-aws-2026",
   "type": "post",
   "title": "The State of the Tech: Quantum Computing on AWS (2026)",
   "url": "https://quantumdev.io/blog/the-state-of-the-tech-quantum-computing-on-aws-2026/",
   "markdownUrl": "https://quantumdev.io/blog/the-state-of-the-tech-quantum-computing-on-aws-2026.md",
   "description": "Where quantum computing stands in 2026: what today's hardware can do, its limits, and how to build hybrid optimization and machine learning workflows on AWS.",
   "published": "2026-06-01",
   "updated": "2026-06-01",
   "author": "Christian Perez",
   "text": "# The State of the Tech: Quantum Computing on AWS (2026)\n\nBy Christian Perez, published 2026-06-01\n\nCanonical: https://quantumdev.io/blog/the-state-of-the-tech-quantum-computing-on-aws-2026/\n\nWhere quantum computing stands in 2026: what today's hardware can do, its limits, and how to build hybrid optimization and machine learning workflows on AWS.\n\nQuantum computing is difficult to assess from the outside.\n\nCoverage tends to overstate its near-term disruption or deep-dive into theory. Both approaches leave a gap where a practitioner needs clarity. What we need is an honest account of current capabilities and a concrete path to building with quantum computing.\n\nThis article is an attempt to address this lack of clarity. Despite quantum computing being in its nascent stages, developers, innovators, and business leaders need to understand this technology's current capabilities as well as what use cases and business opportunities can be addressed.\n\nAmazon Braket, AWS' quantum computing service, is already well documented, so the goal here is not to restate it but to offer a builder's map: what the hardware can genuinely do today, which use cases are crossing from research into early production, where the hard limits sit, and how to get hands-on without standing up a cryogenics lab (the kind of thing you'd only need if you were building the hardware yourself). Our foundational approach will be to treat quantum as another accelerator inside a hybrid system, a coprocessor you reach for on specific structured problems, rather than a replacement for the classical stack you already run. That posture is the most useful one a builder can hold in 2026, and it is why the focus here falls on the two workloads most likely to intersect with your existing work: optimization and quantum machine learning.\n\n## Why run quantum workloads in the cloud instead of a lab?\n\nAmazon Braket is AWS's fully managed quantum computing service. It gives you a single environment to design quantum circuits, test them on simulators, and run them on real quantum hardware from multiple providers, all through a familiar SDK and the AWS console you already use. Rather than buying a machine or managing a lab, you call quantum processors the same way you would call any other AWS service, paying only for what you run.\n\nThat framing matters once you see what the underlying hardware actually is. Quantum machines are incredibly complex. Trapped-ion systems hold individual ionized atoms in electromagnetic fields. Superconducting chips run near absolute zero. Neutral-atom systems arrange hundreds of atoms with optical tweezers. You do not want to own, cool, or calibrate any of this, and with a managed service you never have to.\n\nBraket reached general availability in August 2020, making it one of the earliest multi-vendor quantum services offered by a major cloud provider. The value for a builder on AWS is not just the qubits themselves. It is that the same IAM, VPC, S3, and CloudWatch tooling you already use applies to your quantum tasks, and that several hardware vendors with fundamentally different physics live behind one consistent SDK. You write a circuit once and can target a trapped-ion machine, a superconducting chip, or a simulator by changing one device ARN.\n\nThat multi-vendor access means that no single qubit technology has won. As of mid-2026, Amazon Braket lists hardware from AQT and IonQ (trapped ion), IQM and Rigetti (superconducting), and QuEra (neutral atom), alongside on-demand simulators (AWS). Trapped-ion systems offer all-to-all connectivity and high fidelity but slower gates. Superconducting chips are fast but have fixed, limited connectivity. Neutral-atom arrays scale to hundreds of qubits and excel at analog simulation. Each mode of physics these technologies leverage has different benefits, and being able to test the same algorithm across them is a real engineering advantage as we continue to explore quantum computing use cases.\n\n## What can quantum hardware actually do in 2026?\n\nLet's create an accurate mental model of the machines we are renting but before we dive deeper into the hardware and software, we need to have a consensus on common terminology.\n\nTwo terms will recur throughout. A qubit is the quantum equivalent of a classical bit, but instead of being fixed at 0 or 1 it can hold a combination of both at once, which is what gives quantum machines their potential. This capability to exist in two states simultaneously is what quantum physics refers to as superposition.\n\nWe also have quantum gates. A quantum gate is a single operation that manipulates one or more qubits, the basic building block of a quantum circuit in the same way logic gates build classical ones. Two numbers describe how well that hardware behaves: gate fidelity which is the probability that a gate does what it is supposed to rather than introducing an error, and the error rate, its inverse. Those two ideas, fragile qubits and imperfect gates, explain almost every limit that follows.\n\nNow back to the current state of the tech.\n\nCurrently, we are in what the quantum computing field calls the NISQ era. NISQ stands for Noisy Intermediate-Scale Quantum. This is a term coined by Caltech physicist John Preskill to describe processors that are powerful but not yet error-corrected. These processors hold somewhere between tens and roughly a thousand physical qubits. These processors are also sensitive to their environment, and cannot yet run continuous error correction. The practical constraint is important and worth acknowledging: two-qubit gate fidelities sit around 95 to 99 percent, and with error rates above roughly 0.1 percent per gate, a circuit can execute on the order of 1,000 gates before noise overwhelms the signal. Every algorithm you design for today's hardware lives inside that depth budget.\n\nThe encouraging news is that 2025 was a genuine inflection year. Quantinuum's Helios system, launched in November 2025, packs 98 physical qubits with two-qubit gate fidelity of 99.921 percent and produced 48 fully error-corrected logical qubits at a roughly 2:1 physical-to-logical encoding ratio, an efficiency considered impossible just a few years ago. AWS itself entered the error-correction race with Ocelot, announced in February 2025, an architecture built on cat qubits that AWS says could reduce the resources needed for error correction by up to 90 percent compared with conventional approaches. Harvard's Mikhail Lukin, co-founder of QuEra, now argues recent advances put fault-tolerant machines five to ten years ahead of prior estimates. His group, working with QuEra, has demonstrated error correction below the critical threshold, the point where adding qubits reduces errors rather than introducing them, with useful quantum computing \"in our direct line of sight\" by the end of this decade.\n\nSo what can we as builders and innovators take away from all of this?\n\nFault tolerance is coming faster than the 2020 consensus expected, but it is not here for arbitrary workloads.\n\nWhat we can do right now is extend the usable depth of current machines with error mitigation, which already stretches reliable circuits from hundreds of operations into the thousands and beyond, enough to run genuine experiments. We can deploy quantum-inspired classical algorithms that capture some of the structure of quantum methods and deliver measurable gains on the hardware we already own. And we can build, validate, and operationalize the hybrid quantum-classical workflows that serious teams are already running in production pilots, from AstraZeneca's chemistry work with IonQ, AWS, and NVIDIA to JPMorgan's option-pricing research with IBM.\n\nNone of this is a placeholder for \"real quantum\" later. The orchestration, the cost discipline, the benchmarking against strong classical baselines, and the judgment about which problems are even worth encoding all transfer directly to better hardware. The hardware will cross the threshold in due time. The fluency to use this technology is the part we need to build now.\n\n## Can quantum computing solve optimization problems today?\n\nOptimization is the use case most often pitched to enterprises. Logistics, supply chains, scheduling, traffic flow, satellite placement, and portfolio construction all share a structure: an myriad of possible configurations where classical solvers slow down as the number of solutions increases. As SAS quantum lead Bill Wisotsky puts it, if the solution space is narrow you can solve it classically with ease, but when it is very broad with many possibilities, that is where quantum becomes interesting.\n\nThe ideal algorithm here is QAOA, the Quantum Approximate Optimization Algorithm. It is a hybrid design built specifically for noisy hardware: a quantum circuit prepares a candidate solution state using two alternating operations, a cost layer and a mixing layer. A classical optimizer tunes the circuit parameters across iterations. You map your problem to a cost function, typically a QUBO (Quadratic Unconstrained Binary Optimization) formulation, where each decision becomes a 0-or-1 variable and the quantity you want to minimize is written in terms of those variables. The quantum-classical loop then searches for the lowest-energy configuration, which corresponds to the best solution.\n\nTo make this concrete, here is the heart of a QAOA MaxCut circuit, the canonical graph-partitioning benchmark, written with the Braket SDK:\n\n```python\nfrom braket.circuits import Circuit\n\ndef qaoa_circuit(graph_edges, n_qubits, gammas, betas):\n    circuit = Circuit()\n    n_layers = len(gammas)\n\n    # Start in equal superposition over all candidate solutions\n    for q in range(n_qubits):\n        circuit.h(q)\n\n    for layer in range(n_layers):\n        # Cost layer: penalize edges whose endpoints share a partition\n        for (i, j) in graph_edges:\n            circuit.cnot(i, j)\n            circuit.rz(j, gammas[layer])\n            circuit.cnot(i, j)\n        # Mixer layer: explore neighboring solutions\n        for q in range(n_qubits):\n            circuit.rx(q, 2 * betas[layer])\n\n    return circuit\n```\n\nThat is the entire quantum kernel. The interesting engineering is everything around it: the classical optimizer adjusting gammas and betas (the tunable angles in each layer that the classical loop searches over to improve the solution), the shot budget (how many times you run and measure the circuit, since each run yields one probabilistic sample and you need many to estimate an answer), and the orchestration (the surrounding machinery that runs the quantum-classical loop, manages job submission and results, and keeps cost and runtime under control).\n\nQAOA does not reliably beat strong classical solvers yet. A 2025 analysis in Nature Communications Physics found that noiseless QAOA can outperform the standard Goemans-Williamson semidefinite-programming baseline for MaxCut once you reach six or more layers, but the ACTUAL hardware runs (both error-detected and unencoded) had not yet surpassed that classical baseline. A 2026 survey of industrial logistics came to the same conclusion: current demos are limited in scale, solution quality is usually comparable rather than superior to strong classical heuristics, and performance is constrained by noise, shallow circuits, and the overhead of embedding problems into QUBO form.\n\nSo where is the real value right now? Three places.\n\nFirst, hybrid pipelines where classical methods handle preprocessing and constraint checking and the quantum process addresses the hardest part: the search through a space of possible solutions that grows as the problem scales.\n\nSecond, targeted pilots where classical solvers genuinely hit a wall, such as vehicle routing or complex portfolio optimization.\n\nThird, the quantum-inspired classical algorithms that mimic quantum dynamics on conventional hardware and deliver real gains today while the hardware matures.\n\nResearch is also moving past plain QAOA. Novel methods have shown empirical runtime advantages over state-of-the-art classical optimizers on hard benchmark instances. There is also active work applying deep reinforcement learning to tune QAOA for problems like container-terminal scheduling, beating first-in-first-out and genetic-algorithm baselines under variable conditions.\n\nAs builders and innovators it is important that we learn to stand up the hybrid optimization pattern now, benchmark it honestly against our best classical solvers, and identify what real-life problems we can solve.\n\n## Where does quantum machine learning stand?\n\nQuantum Machine Learning or QML is the use case closest to what AI engineers already do. In QML, the goal is to encode classical data into a quantum state, run it through a parameterized circuit that acts like a neural network, and measure an output. The analogy to a neural network is clear: encoding the data is the input layer. The hidden layer is a stack of quantum gates with tunable settings adjusted during training. Lastly, measuring the qubits at the end is the output.\n\nThere are two main approaches worth knowing.\n\nVariational Quantum Classifiers are the more direct of the two. You build a quantum circuit with tunable parameters, feed in your data, measure a prediction, and train the whole thing end-to-end the way you would train a neural network. You compute how wrong the prediction is, calculate the gradient (the direction to nudge each parameter to reduce that error), and then simply update. The problem is that you cannot inspect a quantum circuit's internals mid-computation the way you can with a classical model, so the usual method of reading off gradients does not apply. The parameter-shift rule solves this elegantly. It computes the exact gradient of a parameter by running the circuit twice, once with that parameter shifted slightly up and once slightly down then taking the difference. This matters because the obvious alternative, finite differences, only approximates the gradient and is highly sensitive to the measurement noise inherent in quantum hardware. The parameter-shift rule, on the other hand, gives a mathematically exact result from the same two runs.\n\nQuantum kernel methods take a different route. Let's first take a look at the classical approach. Many machine learning tasks come down to measuring how similar two data points are. A kernel is exactly that: a function that takes two inputs and returns a single number scoring their similarity. A support vector machine, one of the most reliable classical classifiers, uses these similarity scores to draw the boundary between classes (for example separating spam from not-spam). The trick that makes kernels powerful is that they implicitly compare data not in its raw form but in a transformed, higher-dimensional representation called a feature space, where classes that looked tangled together may become separable.\n\nThe quantum version keeps the classical support vector machine but uses the quantum computer for one job only: computing the kernel. You encode each data point into a quantum state through a feature map (the circuit that turns classical data into the high-dimensional space of the quantum system), then measure how much two such states overlap to get their similarity score. The appeal is that a quantum feature map can reach a feature space that grows exponentially with the number of qubits, one so large that no efficient classical method can represent it. In principle, that gives access to similarity patterns, and therefore class boundaries, that classical kernels cannot practically reach.\n\nOn Braket, the practical entry point for both is PennyLane, the differentiable-programming framework integrated into the service.\n\nSwitching from a local simulator to managed hardware is a simple a one-line change:\n\n```python\nimport pennylane as qml\n\n# Prototype locally, then point the same code at managed simulators or QPUs\ndev = qml.device(\"braket.local.qubit\", wires=4)\n\n@qml.qnode(dev)\ndef circuit(params, x):\n    for i in range(4):\n        qml.RY(x[i], wires=i)          # encode features as rotations\n    qml.StronglyEntanglingLayers(params, wires=range(4))  # trainable layers\n    return qml.expval(qml.PauliZ(0))   # measure to get a prediction\n```\n\nAs of 2025, practical demonstrations of quantum advantage in supervised learning remain limited. They are mostly confined to synthetic benchmarks or small datasets, with no reproducible advantage on real-world classical data. On top of that, training quantum models runs into hard, well-documented failure modes: gradients that flatten out as circuits grow, and similarity scores that blur together until the model can no longer tell data points apart.\n\nThese are not reasons to give up. They are reasons to keep building deliberately.\n\nThe builder's approach to QML should treat the tech as a research-grade capability we develop in-house, validate against classical baselines, and aim first at quantum or physics-based data.\n\n## How do you take a quantum workload to production on AWS?\n\nThe most overlooked truth about quantum computing on AWS is that the challenging part is not the circuit. Orchestrating the quantum-classical loop, logging metrics, checkpointing long-running optimizations, and packaging it all so it survives interruptions and scales; that is the challenge.\n\nThis is exactly what Braket Hybrid Jobs exists for. While preparing this article I created an open-source curriculum that builds Braket workflows from first principles up to production patterns, including QAOA and QML implemented as Hybrid Jobs with the operational scaffolding intact.\n\nThere are a few patterns from the curriculum worth highlighting, and checkpointing is the clearest.\n\nCheckpointing is a non-negotiable for variational algorithms. A training loop that runs for many iterations against managed hardware needs to resume from its last good state rather than restart, which the Jobs API supports directly:\n\n```python\nfrom braket.jobs import save_job_checkpoint, load_job_checkpoint\n\ncheckpoint = load_job_checkpoint()\nif checkpoint:\n    params = checkpoint[\"params\"]      # resume an interrupted training run\n    start_epoch = checkpoint[\"epoch\"]\nelse:\n    params = initialize_params()       # fresh start\n    start_epoch = 0\n```\n\nIt is imperative that we apply cloud computing first-principles (validation, checkpointing, observability, and cost control) to quantum computing. This is why a cloud and AI engineer is well positioned to leverage this modern technology and begin to explore its capabilities.\n\n## The Bottom Line\n\nQuantum computing in 2026 is neither the imminent revolution nor a mirage. It is an emerging accelerator with a small but growing set of problems where it is becoming genuinely interesting, despite the real and well-understood limits.\n\nThe builders who will set themselves apart are the ones building these hybrid workflows now. With Amazon Braket, AWS enables us to swap hardware with one line, validate against classical baselines, and treat the skill itself as the asset. The speedups and improvement in quantum technology will come on their own schedule. The fluency we build while waiting is entirely within our control.\n\nThis article was first published on [AWS Builder Center](https://builder.aws.com/content/3EVVktCFbwlupNiJCkNvUp8Wx1z/the-state-of-the-tech-quantum-computing-on-aws-2026) on 1 June 2026.\n"
  }
 ]
}
