Google Stitch Loop: The Agent That Opens PRs After You Ship
Google Labs opened the Stitch Loop waitlist on November 18, 2026. It watches live metrics and opens GitHub pull requests. Nothing merges until you review.

Google Labs opened the Stitch Loop waitlist on November 18, 2026. It watches live metrics and opens GitHub pull requests. Nothing merges until you review.

On November 18, 2026, Justin Toupin, Director of Product Management at Google Labs, published Introducing Stitch Loop ↗. Google calls it an experiment and "a new direction for Stitch": a long-running product agent that connects to a codebase and to live metrics and keeps working 24/7 until the goals are met. Google says millions of designers, developers, and founders already use Stitch. TestingCatalog ↗ (Alexey Shabanov, November 18) and BriefPixel ↗ (November 18) covered the launch. Google press image (Stitch Loop launch post, Nov 18, 2026), used for news reporting; credit Image: Google. We have not used Stitch Loop.
Setup, in Google's post: connect GitHub and product context, including logs, user behavior, and feedback, through more than 180 integrations, custom MCP servers, or local CLI uploads.
Each loop runs four steps. Suggestions are grounded in evidence from the product. Solutions come as one or more options, with tradeoffs, scoping, and visual design options where Google says they apply. Build writes the code in an isolated cloud sandbox and opens a GitHub pull request. When relevant, Google says the pull request contains screenshots and videos. Reflect monitors live product data, learns from the results, and keeps watching after the goal is met. Google's line: "Nothing merges until you review the PR."
| Step | Result |
|---|---|
| Suggest | Evidence |
| Solve | Tradeoffs |
| Build | GitHub PR |
| Reflect | Live data |
Three default loops ship with it: cybersecurity, latency, and accessibility. A team can add a custom loop, and Google's example is checkout conversion. The security loop pairs Gemini 4 Carbon's cyber capabilities with CodeMender, which Google describes as Google DeepMind's code-security agent. It flags issues with evidence and opens fix pull requests. Gemini 4 Carbon is that model's note. The same-day sibling, a different product, is Gemini 4 Neon.
Google says Stitch Loop also works as a service layer for coding agents such as Google Antigravity, through the Stitch CLI and MCP. In Slack, @stitch sends daily insight digests and creates loops. Our Antigravity note covers that agent. The MCP tutorial is the protocol background.
The waitlist is open globally for all users at stitch.google.com/loop/waitlist ↗. Eligible Google AI Ultra users get priority access. BriefPixel says Google's announcement gives no general release date and no standalone price, and that joining the waitlist is not immediate access. The launch post matches that: it names the waitlist and the Ultra priority, and it does not name a price or a release date.
TestingCatalog says it tested Loop for weeks before launch on TestingCatalog.com. It says the agent "made many small tweaks to the theme" and suggested accessibility and compliance improvements. It says that pre-launch run used Gemini 3.8 Flash. It calls Loop "a rebirth of Jules". Those sentences are TestingCatalog's account of its own site. They are not a result from this desk, and the pre-launch model it names is not the Carbon pairing Google describes for the security loop now.
"a product agent built for outcomes, not just PRs." — Stitch by Google ↗ (@stitchbygoogle)
"Stitch Loop watching live metrics and opening PRs after launch is a wild amount of trust to hand a model." — fity.eth ↗ (@Fityeth)
"If you missed Jules, you won't miss it anymore" — Rick De Oliveira ↗ (@r1ckdeoliveira)
"I thought Google Stitch was dead." — sunimod_xø ↗ (@0xdominus)
Reddit was not reachable for this story.
This is a waitlist story for a small product team that already has traffic and a GitHub repo. Google has not published a price or a general release date. BriefPixel's reading matches the launch post: the form is not a login. Ultra priority is a queue position Google describes, not a ship date.
The human pull-request gate is the safeguard Google states. "Nothing merges until you review the PR." fity.eth's sentence is the other half. Handing a model live metrics, and the ability to open pull requests, is a trust decision. Connecting production logs and user behavior is a data-governance decision before it is a feature. Review the diff the way you review any other agent's patch. The AI code review tools guide is the human side of that gate.
Coding agents that stop at merge are a different job. Google's own post says a coding agent's job ends when the pull request merges, and that the production work starts after that. Devin Bots is Cognition's persistent-agent product. Cursor Build is Cursor's app builder. Claude Code and Codex sit in the best AI for coding guide. Antigravity is Google's coding agent, and Google says Stitch Loop can feed it product context. Rick De Oliveira's Jules line and TestingCatalog's "a rebirth of Jules" make the same comparison. sunimod's "I thought Google Stitch was dead" is the reputation this launch is answering. Stitch's own line, "a product agent built for outcomes, not just PRs," is the pitch. The pull request is still how the work arrives.
Verdict: Join the waitlist if a real repository and real traffic make a monitored loop useful, and keep the merge on a person. Do not treat the form as access, a price, or a date.