BetaEditorial Listing

Stitch Loop

Google · Stitch Loop: Google Labs' Long-Running Product Agent That Turns Live Metrics into Pull Requests

Open Stitch Loop

Stitch Loop is a long-running product agent from Google Labs that connects to your GitHub codebase and live product data, proposes changes toward goals such as security, latency, accessibility or checkout conversion, builds the chosen fix in a cloud sandbox and opens a pull request for you to review. It is aimed at lean product and engineering teams that have shipped and need ongoing improvement work done, and it is currently waitlist-only.

PricingCustom
Setupmedium
Runs onWeb
Open sourceNo
CategoryCoding
Product AgentContinuous OptimizationPull RequestsGitHubProduct AnalyticsApplication SecurityPerformanceAccessibilityConversion OptimizationMCPSlackGoogle LabsGeminiWaitlist

Best for

Lean product and engineering teams with a live product on GitHub and measurable signals in logs, analytics and feedback, who want an agent to keep proposing and building reviewed fixes for security, performance, accessibility or conversion after launch

Not ideal for

Teams that need a supported, generally available product with published pricing and data terms today, code hosted outside GitHub, projects without live usage data to optimize against, and anyone who wants a task-on-demand coding agent rather than a goal-driven one

Who it's for

Lean product, engineering and design teams maintaining and improving a live software product on GitHub

Capabilities

  • Connects to GitHub plus product context such as logs, user behavior and feedback through 180+ integrations, custom MCP servers or local CLI uploads
  • Runs long-lived, goal-based loops that work toward a measurable outcome 24/7 rather than a single task
  • Three built-in loops for cybersecurity, latency and accessibility, plus custom loops for your own goals such as checkout conversion
  • Suggestions step surfaces opportunities grounded in evidence from your product data
  • Solutions step offers one or more options per suggestion, each with tradeoffs and scoping, and visual design options where applicable
  • Build step writes the chosen solution in an isolated cloud sandbox and opens a GitHub pull request, including screenshots and videos of the work when relevant
  • Reflect step monitors live product data, learns from positive and negative results, and keeps watching after a goal is met in case the metric slips
  • Security loop pairs Gemini 4 Carbon's cybersecurity capabilities with Google DeepMind's CodeMender to scan the codebase, flag each issue with evidence and propose a fix as a PR
  • Acts as a product-context layer for other coding agents, such as Google Antigravity, through the Stitch CLI and MCP
  • @stitch Slack app for daily insight digests and for creating new loops

Limitations

  • Waitlist-only Google Labs experiment announced November 18, 2026; eligible Google AI Ultra subscribers get priority access
  • No pricing has been published
  • No public documentation, full integration list, data-retention or model-training terms for Loop were available at launch
  • Changes reach your product only as GitHub pull requests; GitHub is the only code host named in the announcement
  • Every change needs human review before it merges, so someone on the team still has to read and approve each PR
  • Its suggestions and the Reflect step depend on connecting meaningful live product data such as logs, analytics and feedback, which takes setup beyond linking a repository
  • No self-hosted deployment is documented

Use cases

  • Running a standing security loop that scans the codebase and opens fix PRs for flagged vulnerabilities
  • Setting a latency goal for a slow flow such as checkout and reviewing the PRs Loop proposes until the metric improves
  • Working through accessibility issues in a shipped web product as a continuous loop instead of a one-off audit
  • Creating a custom loop to improve a conversion metric such as signup or checkout completion, with design options to choose from
  • Feeding persistent product context into Google Antigravity or another coding agent through the Stitch CLI and MCP
  • Getting a daily digest of product insights in Slack and starting new loops with @stitch

Our take

Stitch Loop's idea is that the hard part of a product is what happens after the coding agent's PR merges: vulnerabilities, slow flows and drop-offs that only show up in production data. Instead of waiting for a ticket, it holds a goal, reads the evidence from your logs, analytics and feedback, offers options with tradeoffs, builds the chosen one in a sandbox and keeps measuring. That is a different job from Devin, Codex or Factory, which execute work you assign, and every change Loop builds itself reaches your repository only as a pull request a person must review. The catch is that almost everything beyond the launch post is still unknown: there is no pricing, no documentation, no list of the 180+ integrations and no data terms, and it is a Labs experiment behind a waitlist. Teams with a live GitHub product and real metrics to optimize should join the waitlist and evaluate it on one loop, ideally security or latency where results are easy to verify, before relying on it.

Who should use it

Small product teams and founders who have shipped and lack people for ongoing security, performance and accessibility work; product engineers who want suggestions tied to live metrics rather than a backlog; and eligible Google AI Ultra subscribers, who get priority on the waitlist. Teams already using Antigravity or Slack can also connect Loop to those tools.

Who should skip it

Organizations that need procurement-ready pricing, documentation and data-processing terms now; teams whose code is not on GitHub; pre-launch projects with no live data; and anyone who wants an agent to execute specific tickets on demand, where a task-based coding agent is a better fit.

Strengths

  • Works from live product data toward a measurable goal and keeps monitoring after the goal is met, rather than stopping at a merged PR
  • Built-in loops for security, latency and accessibility, plus custom goals
  • Offers several solutions with tradeoffs and design options before building anything
  • Builds in an isolated cloud sandbox and never merges without human review of the pull request
  • Security loop draws on Google DeepMind's CodeMender
  • Can feed product context to other coding agents through the Stitch CLI and MCP, and works from Slack

Weaknesses

  • Waitlist-only experiment with no published pricing or general availability date
  • No public documentation, integration list or data-handling terms yet
  • Only GitHub is named as a code host
  • Needs live logs, analytics and feedback connected to be useful, not just a repository
  • As a Google Labs experiment, its scope and availability may change

Stitch Loop pricing

Waitlist

Not published

  • Waitlist open to all users globally since November 18, 2026
  • Priority access for eligible Google AI Ultra subscribers

Note: Google has not published pricing for Stitch Loop. Access is through the waitlist at stitch.google.com/loop/waitlist, and Google says eligible Google AI Ultra subscribers get priority; it has not said whether Loop will be included in Google AI plans or billed separately.

Technical specs

Available models

Gemini 4 Carbon (named by Google for the security loop, alongside the CodeMender code-security agent)

Where Stitch Loop excels

Continuous vulnerability remediation

The built-in security loop scans the codebase with Gemini 4 Carbon and CodeMender, flags each issue with evidence and opens a fix PR, which suits small teams without a dedicated security engineer.

Performance and conversion goals

A latency loop or a custom checkout-conversion loop proposes options, builds the chosen change and watches the live metric afterwards, so the team spends its time choosing among options and reviewing PRs rather than diagnosing the problem from scratch.

Product context for an existing coding agent

Teams already using Antigravity or another agent can pull Loop's persistent product context through the Stitch CLI and MCP, so their agent works from production evidence rather than only the code.

Stitch Loop vs. competitors

Stitch Loop vs. Devin

Devin is Cognition's autonomous software engineer: you assign tickets from Slack, Linear, Jira or its app and it opens pull requests from its own cloud VM. Stitch Loop also ends in pull requests, but it decides what to work on from live product data and a standing goal, and keeps monitoring the metric afterwards.

Stitch Loop vs. CodeMender

CodeMender is a code-security agent, originally from Google DeepMind and now offered by Google Cloud in a limited preview, that finds, verifies and patches vulnerabilities; Stitch Loop's built-in security loop pairs it with Gemini 4 Carbon. Choose CodeMender for dedicated security remediation, and Loop if security is one of several product goals you want handled by the same agent.

Stitch Loop vs. Factory Droid

Factory's Droids cover the software delivery lifecycle (planning, coding, review, testing) from a desktop app, CLI, IDEs, CI and Slack for teams assigning work. Stitch Loop focuses on the post-launch phase, choosing improvements from production signals rather than from assigned tasks.

Stitch Loop vs. OpenAI Codex

Codex reads, edits, runs and reviews code in your repository from a CLI, IDE or cloud tasks you start. Stitch Loop is not a general coding agent you prompt for tasks; it runs long-lived, goal-based loops and can supply product context to coding agents through the Stitch CLI and MCP.

Frequently asked questions

What is Stitch Loop?

Stitch Loop is a long-running product agent from Google Labs, announced on November 18, 2026. It connects to your GitHub codebase and product context such as logs, user behavior and feedback, runs goal-based loops for security, latency, accessibility or your own goals, builds proposed fixes in an isolated cloud sandbox and opens pull requests for you to review.

How do I get access to Stitch Loop, and how much does it cost?

Access is through a waitlist at stitch.google.com/loop/waitlist, open to all users globally, with priority for eligible Google AI Ultra subscribers. Google has not published pricing.

Does Stitch Loop change my code without approval?

No. Google says nothing merges until you review the pull request. Loop writes code in an isolated cloud sandbox and submits it as a GitHub PR, with screenshots and videos when relevant.

What data sources can Stitch Loop connect to?

Google says you connect GitHub and product context such as logs, user behavior and feedback through over 180 integrations, custom MCP servers or local CLI uploads. The full list of integrations has not been published.

How does Stitch Loop relate to Stitch and CodeMender?

Stitch is Google Labs' AI tool for generating app interfaces, and Google describes Loop as part of a new direction for Stitch covering products from ideation through continuous optimization. Loop's built-in security loop pairs Gemini 4 Carbon with CodeMender, Google DeepMind's code-security agent.

Can I use Stitch Loop with other coding agents?

Yes. Google says coding agents such as Google Antigravity can pull persistent product context from Stitch Loop through the Stitch CLI and MCP, and an @stitch Slack app sends daily insight digests and creates new loops.

Integrations & fit

GitHubSlackMCP serversStitch CLIGoogle Antigravity
Good fit forSolo / individual, Startup / small team
Pricing modelCustom· Contact for pricing
See pricing on Stitch Loop →

Alternatives to consider

About Stitch Loop

Most coding agents in this directory start from a task you give them and stop when their pull request is opened or merged. Stitch Loop starts from a goal and keeps going after launch. Google Labs announced it on November 18, 2026 as an experiment and part of a new direction for Stitch, its AI design tool. You connect GitHub and your product context, such as logs, user behavior and feedback, through more than 180 integrations, custom MCP servers or local CLI uploads. Loop then runs evergreen loops, each aimed at a measurable outcome. Three come built in, for cybersecurity, latency and accessibility, and you can add your own, such as improving checkout conversion. Every loop has four steps. Suggestions surface opportunities grounded in evidence from your product. Solutions give one or more ways to act on each suggestion, with tradeoffs, scoping and, where relevant, visual design options. Build writes the chosen solution in an isolated cloud sandbox and opens a GitHub pull request, with screenshots and videos where relevant. Reflect monitors live data, learns from what worked, and keeps watching after the goal is met in case the metric slips. Nothing merges until a person reviews the pull request. The security loop pairs Gemini 4 Carbon's security capabilities with CodeMender, Google DeepMind's code-security agent, and opens a PR with a proposed fix for each issue it flags. Loop can also act as a context layer for other coding agents such as Google Antigravity, which can pull product context from it through the Stitch CLI and MCP, and an @stitch Slack app sends daily insight digests and creates new loops. The tradeoffs are significant for now: access is through a waitlist (eligible Google AI Ultra subscribers get priority), Google has published no pricing, documentation, list of the 180+ integrations, or data-handling terms for Loop, and it is a Google Labs experiment rather than a generally available product. Its value also depends on wiring meaningful live metrics into it, which is real setup work.

Updates from Stitch Loop

LaunchGoogle Labs announces Stitch Loop and opens the waitlist

Google Labs introduced Stitch Loop on November 18, 2026 as an experiment and a new direction for Stitch: a long-running product agent that connects to GitHub and live product data, runs goal-based loops for security, latency, accessibility and custom goals, and opens pull requests for human review. The waitlist opened to all users globally the same day, with priority access for eligible Google AI Ultra subscribers.

Are you the founder? Claim this listing →