LATEST
AI Tools

Google Stitch Loop: Always-On AI Agent, Waitlist Open, No Price Yet

Updated November 20, 2026
15 min read
Editor's Note: This article was human-reviewed for accuracy
Team of developers reviewing code together on laptops in a modern office

Quick Answer & Key Takeaways

  • • Google Labs launched Stitch Loop on November 18, 2026, an always-on product agent that watches your codebase and live metrics, then writes fixes and opens pull requests for your review.
  • • The waitlist is open worldwide, with eligible Google AI Ultra subscribers getting priority access, but Google has published no price and no general-availability date.
  • • Its security loop pairs Gemini 4 Carbon with CodeMender, Google DeepMind's code-security agent, while latency and accessibility loops cover the other two built-in tracks.

Coding agents have a blind spot, and Google just named it. They build fast and ship fast, then their job ends the moment a pull request merges. The work that comes after, patching a vulnerability, speeding up a slow checkout, figuring out why signups are dropping, is still mostly manual, and for a small team it never really ends. On November 18, 2026, Google Labs introduced Stitch Loop, a long-running product agent that connects to your codebase and your live metrics and keeps working on your product around the clock until your goals are met.

It is an experiment, not a product you can buy. The waitlist opened the same day, worldwide, with eligible Google AI Ultra subscribers moved to the front of the line. Google has published no price for Loop itself and no date for wider access. That is the honest starting point for this article: everything below comes from Google's launch post and the surrounding documentation, and where Google has said nothing, this article says so.

What Stitch Loop actually does

Stitch Loop is the second half of Stitch. The original Stitch, launched at Google I/O in May 2025, turns a text prompt, a sketch, or an image into a working interface and front-end code. Google says millions of designers, developers, and founders already use it. Loop picks up where Stitch stops: after the product is live.

You connect Loop to your GitHub repository and to the context around your product, logs, user behavior, and feedback, through more than 180 integrations, custom MCP servers, or uploads from a local command-line tool. Grounded in that context, Loop runs what Google calls evergreen loops, continuous cycles of work aimed at measurable outcomes you set, such as fewer security issues, faster pages, or a higher checkout conversion rate. It keeps going until the goal is met, and then it keeps watching in case the metric slips again.

The important design choice is that Loop does not merge anything on its own. It writes code in an isolated cloud sandbox and opens a GitHub pull request for your approval. Nothing lands in your codebase until a human reviews it. That single rule is what separates Loop from the fully autonomous agents that have made headlines this year, and it is the first thing to check in any tool that touches production code.

How it works: four steps, repeated

Every loop runs through the same four stages. Google describes them as Suggestions, Solutions, Build, and Reflect, and they repeat for as long as the loop is active.

1. Suggestions

Loop surfaces opportunities to move toward your goal, grounded in evidence from your product. That means a suggestion should come with a reason tied to something real, a slow endpoint in your logs, a drop-off in your analytics, a vulnerability in your code, rather than a generic recommendation.

2. Solutions

Every suggestion comes with one or more proposed solutions, each with its tradeoffs and scoping, and visual design options where they apply. This is the step where Loop is supposed to show its working: what it would change, what it would cost in effort, and what could go wrong.

3. Build

Once you pick a direction, Loop writes the code in an isolated cloud sandbox and opens a GitHub pull request for your approval. Where relevant, the pull request includes screenshots and videos of the work Loop did, so you can see the change running before you read the diff.

4. Reflect

Loop monitors live product data as it works toward your goal, learning from every positive or negative result to sharpen the next round of suggestions. Once the goal is met, it keeps watching in case the metric slips. This is the step that makes Loop different from a one-off coding session: the agent stays attached to the outcome, not just the task.

The three built-in loops

Loop ships with three ready-made tracks, and you can add your own. The built-in three cover the jobs small teams most often defer.

LoopWhat it watchesWhat it does
SecurityYour codebase, scanned for vulnerabilitiesFlags each issue with evidence and opens a pull request with a proposed fix
LatencyLive performance data from your productFinds slow paths and proposes fixes aimed at faster responses
AccessibilityYour interface against accessibility standardsSurfaces WCAG issues and proposes fixes for them

Custom loops cover anything else you can measure. Google's example is improving checkout conversion, which is a telling choice: it frames Loop as a product tool, not just a developer tool. A conversion loop would watch funnel metrics, propose changes to the checkout flow, and test them the same way the security loop tests a patch. Whether that works in practice depends on how well Loop can connect a metric to a specific code change, and no independent test has measured that yet.

The security loop: Gemini 4 Carbon meets CodeMender

Google gives the most detail about the security loop, and it is the strongest part of the announcement. The loop pairs the cybersecurity capabilities of Gemini 4 Carbon, Google's frontier model, with CodeMender, Google DeepMind's autonomous agent for improving code security.

CodeMender is not new. DeepMind introduced it in October 2025 as a research project, and Google Cloud put it into public preview on July 21, 2026, on the Gemini Enterprise Agent Platform. In its research phase, DeepMind said CodeMender produced 72 verified patches that were upstreamed to major open-source projects under human oversight. Google says hundreds of customers have used it since the preview opened. It scans and remediates vulnerabilities across C, C++, Go, Java, Python, Ruby, Rust, and TypeScript or JavaScript, plus frameworks including Django, Flask, React, Spring Boot, and Express.

CodeMender's method matters more than its coverage list. Rather than only flagging a possible flaw, it builds and runs proof-of-concept exploits in a customer-managed sandbox to check whether a vulnerability is actually exploitable, then writes a tested fix. Google's early evaluation says it has the potential to cut vulnerability discovery time by more than 95 percent, a figure that comes from Google's own testing and has no independent verification yet. Treat it as a direction, not a result.

One caveat carries over from CodeMender into Loop. CodeMender is still in preview, not generally available, and a specialized security-tuned model called Gemini 3.5 Flash Cyber is in a limited-access pilot restricted to governments and trusted partners. The security loop in Stitch Loop therefore rests on a preview-stage agent. That does not make it useless, but it does mean the loop's strongest feature is the one with the least settled status.

What Stitch Loop costs

There is no published price for Stitch Loop, because there is nothing to buy yet. Google has not announced a price, a pricing model, or a general-availability date. The waitlist is the only access route, and joining it costs nothing. The honest price story is therefore about the two things that do have prices: the subscription that gets you priority access, and the model that powers the security loop.

Eligible Google AI Ultra subscribers get priority access to the waitlist. Ultra is Google's top consumer AI subscription. Google restructured it in 2026: the current US tiers run $99.99 a month for 5x Pro usage limits and $199.99 a month for 20x, down from the $249.99 launch price in May 2025. At November 20, 2026 market rates (1 USD = £0.76, C$1.43, approximate), that is about £76 or C$143 a month for the 5x tier and about £152 or C$286 for the 20x tier. Google AI Pro, the tier below Ultra, costs $19.99 a month or $199.99 a year, about £15 or C$29 a month at the same rates. Pro subscribers do not get Loop priority access, so the cheapest route to the front of the queue is the $99.99 Ultra tier, and even that only buys priority, not admission. Google says eligible Ultra users get priority access, not guaranteed access, and it has not said how many people will come off the waitlist or when.

The original Stitch, by contrast, is free with a Google account, according to current reviews of the tool, with monthly generation limits that differ between its standard and experimental modes. Loop is a separate experiment with no announced pricing, so do not assume it will follow Stitch's free model when it opens up.

ItemPriceStatus
Stitch LoopNot publishedWaitlist only, opened November 18, 2026
Google AI Ultra (5x Pro limits)$99.99 a month (about £76, C$143)Priority waitlist access for eligible subscribers
Google AI Ultra (20x Pro limits)$199.99 a month (about £152, C$286)Priority waitlist access for eligible subscribers
Google AI Pro$19.99 a month or $199.99 a year (about £15, C$29 a month)No Loop priority access
Stitch (original design tool)Free with a Google account, monthly generation limitsGenerally available through Google Labs

The model behind the security loop does have a public price, and it is about to change. Gemini 4 Carbon costs $2 per million input tokens and $10 per million output tokens through November 30, 2026, then rises to $4 and $20 from December 1. Cached input costs $0.10 per million tokens now and $0.20 after the change. Those are API rates for developers calling the model directly, not a price for Loop, but they are the closest published number to what Loop's security work will cost Google to run, and they set a floor for whatever Loop eventually charges. If Loop ever bills by usage, Carbon's December price rise is the rate that will apply.

How to join the waitlist

Joining takes minutes, and there is no way to skip the queue except the Ultra priority lane. The steps below come from Google's launch post.

  1. Go to stitch.google.com/loop/waitlist and sign in with your Google account.
  2. Join the waitlist. It is open to all users globally, with no regional restriction stated in the announcement.
  3. If you hold an eligible Google AI Ultra subscription, your request gets priority access. Check that your subscription is active before joining, since Google's wording is "eligible" Ultra users, not all of them.
  4. Share feedback in the Google Labs Discord while you wait. Google points waitlisted users there, which suggests early access will be shaped by the feedback it receives.

Do not expect a confirmation date. Google has given no timeline for invitations, and waitlist-only launches from Google Labs have historically taken weeks to months to reach general users. If you need a maintenance agent now, the waitlist is not a plan.

Stitch Loop vs the alternatives

Loop is not the first agent that works on your codebase without being asked. The difference is what it watches and when it works. Most coding agents start from a prompt; Loop starts from live product data. That puts it next to three tools this site has already covered, each of which approaches the same problem from a different angle.

ToolWhat it doesWho approves the workAccess and price
Google Stitch LoopContinuously improves a live product: security, latency, accessibility, custom goalsYou, on every pull requestWaitlist only; no price published; Ultra subscribers get priority
Devin BotsSpecialist coding agents that run inside a team's workflow, with a manager agent coordinating themYou, through review and the Bots' own reportingEarly access; Bots included from Max at $200 a month
Google Antigravity /teamworkTeams of agents that plan, code, and verify large projects, including formal proofsYou, through the Antigravity workflowBundled with Google AI plans, from the free tier up
OpenAI DaybreakPen-tests and patches code, built around security researchHuman approval is built into the workflowNo published price; research preview and enterprise pilots

The honest comparison is that Loop and Antigravity are complementary, not rivals. Google says so itself: Loop can act as a service layer for coding agents like Antigravity, feeding them persistent product context through the Stitch CLI and MCP. Antigravity builds the thing; Loop keeps it healthy after launch. Devin Bots sit closer to Loop's territory, since Bots also run continuously inside a team's workflow, but they are organized around coding tasks and managed by a coordinator agent, while Loop is organized around product outcomes measured in live data. Daybreak overlaps with Loop's security loop only, and it is a security product first, not a general product agent.

None of these four has an independent head-to-head test behind it yet. Loop has no independent test at all, since nobody outside Google has used it. Any ranking between them today would be guesswork, and this article does not offer one.

The catches

Five gaps matter more than the feature list, and Google's own announcement leaves all five open.

No one can use it yet. Loop is waitlist-only, with no general-availability date. Everything in this article describes a product Google has announced, not one anyone outside the company has used. Access has not opened to anyone yet, because the waitlist is the only route in.

No price exists. Google has published no price, no pricing model, and no statement about whether Loop will be free, bundled into an existing plan, or billed separately. The only price in the story is the Ultra subscription that buys priority access, and priority is not admission.

The security engine is still in preview. CodeMender, which powers the security loop, has been in public preview since July 2026. Its headline figures, 72 verified patches upstreamed and a potential 95 percent cut in discovery time, come from Google's own reporting. There is no independent evaluation of either number.

Priority access is narrower than it sounds. Google promises priority to "eligible" Google AI Ultra subscribers. It has not defined eligibility, and Ultra itself is not available in every country. A subscriber outside the eligible set gets the same waitlist as everyone else.

Google Labs experiments change. Loop is explicitly an experiment from Google Labs, the same label Stitch carries. Google has not said what happens to your loops, your product context, or your connected integrations if the experiment changes direction or shuts down. Teams that connect a production codebase to an experimental agent should plan for that possibility before they connect anything.

Common mistakes

If you join the waitlist, three mistakes are easy to make and worth avoiding.

Treating the waitlist as a launch. Joining the waitlist is not access, and Ultra priority is not a guarantee. Do not plan a maintenance schedule around a tool you cannot open yet.

Connecting production data before reading the terms. Loop needs your GitHub repository and product context, logs, user behavior, and feedback, to do its job. That is a broad grant of access to an experimental product. Read what Google says about data handling for Loop before you connect a live codebase, and start with a non-production repository if you get in early.

Assuming the security loop replaces your own review. Loop's rule is that nothing merges until you review the pull request. That rule only protects you if the review is real. A fix that looks right and passes its own tests can still be wrong, and an agent that writes fixes around the clock can generate more pull requests than a small team can review properly. The review queue is the bottleneck, not the agent.

Frequently asked questions

Is Google Stitch Loop free?

Google has not published a price for Stitch Loop, and it has not said whether it will be free, bundled, or billed separately. Joining the waitlist is free. The only related price is Google AI Ultra, which costs $99.99 a month for the 5x tier and $199.99 for the 20x tier in the US, and which gives eligible subscribers priority access to the waitlist, not free access to Loop itself.

How is Stitch Loop different from Stitch?

Stitch designs interfaces. It turns a prompt, sketch, or image into UI screens and front-end code, and it launched at Google I/O in May 2025. Stitch Loop maintains products that are already live. It watches your codebase and live metrics, finds problems, and opens pull requests with fixes. Stitch is for the start of a product's life; Loop is for everything after launch.

When will Stitch Loop be available?

Google has not announced a general-availability date. The global waitlist opened on November 18, 2026, with priority access for eligible Google AI Ultra subscribers. Google has not said how many people will be invited or when invitations will go out.

Does Stitch Loop merge code by itself?

No. Google states that nothing merges until you review the pull request. Loop writes code in an isolated cloud sandbox and opens a GitHub pull request, with screenshots and videos of its work where relevant, and a human approves or rejects it. That is the product's central safety rule.

What does Stitch Loop need access to?

Your GitHub repository and product context: logs, user behavior, and feedback. You connect these through more than 180 integrations, custom MCP servers, or local CLI uploads. That is a broad grant of access, and it is worth reading Google's data-handling terms for Loop before connecting a production codebase.

What is CodeMender?

CodeMender is Google DeepMind's autonomous code-security agent. DeepMind introduced it in October 2025 as a research project, and Google Cloud put it into public preview on July 21, 2026. It finds vulnerabilities, builds proof-of-concept exploits in a sandbox to check whether they are real, and writes tested fixes. It powers the security loop inside Stitch Loop, paired with Gemini 4 Carbon.

The bottom line

Stitch Loop is a good idea with no price, no date, and no independent test behind it yet. The idea is the strong part: coding agents stop when code ships, and the work that matters most for a small team starts after that. A loop that watches live metrics, proposes fixes, and files pull requests for human review is a sensible answer to that gap, and the security loop has a real engine behind it in CodeMender. The weak part is everything around the idea: waitlist-only access, an unpublished price, a preview-stage security agent, and no evidence yet that the loops find real problems at a useful rate. Join the waitlist if the pitch fits your team, and keep your current maintenance process running while you wait. If you want a coding agent you can use today, read our reviews of Google Antigravity /teamwork and Devin Bots instead.

Sources

15 min read
36 views
edit_note Written by

Ratul Ahamed

Founder & Editor-in-Chief

Ratul Ahamed is the founder and editor of AI Core Updates, an independent site covering AI tools, models and pricing. He is a web developer based in Bangladesh and also runs All In One Tools, a free collection of browser-based utilities. Contact: [email protected]

Read all stories by Ratul Ahamed arrow_forward
verified_user Human-Reviewed Editorial

forum Comments (0)

Leave a Reply

chat_bubble_outline

No comments yet. Be the first to join the conversation!

Weekly AI Digest

Get our weekly AI tools digest

One email a week analyzing new AI tools, model updates, and practical workflows. Tested and summarized with zero fluff. Unsubscribe any time.

bolt
Hands-on AI Tool Reviews
Pricing math, limits, and alternatives.
verified_user
Strict Anti-Spam Policy
1-click unsubscribe at any time.
mark_email_unread

Subscribe to Digest

Sent weekly • No spam

person_outline
mail_outline

No spam ever. You can unsubscribe with a single click at any time.

cookie

Cookie Settings

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. Learn more