Stitch Loop
Intro
Stitch Loop is a long-running product-agent experiment from Google Labs, announced on 18 November 2026, for improving a product after it has already shipped. The official post Introducing Stitch Loop describes it as an agent that connects to your codebase and live metrics, finds what needs fixing, and keeps working around the clock until the goals you set are met.
Justin Toupin, Director of Product Management at Google Labs, presents it as part of a new direction for Stitch. Teams already use Stitch to turn ideas into working interfaces quickly; Stitch Loop extends that work into continuous optimization after launch. Google places the two on one product path, from ideation through ongoing improvement.
This entry records the newly announced experiment name Stitch Loop. It is not the same term as the already filed Stitch AI, and it is not a new name for Gemini 4 Carbon or Google DeepMind’s CodeMender, which appear only in the description of Loop’s security cycle.
Advertisement
Explain
Google’s stated problem is that coding agents can help you ship features and merge pull requests quickly, but their job often ends when code reaches production. After launch, teams still have to patch vulnerabilities, speed up checkout, and diagnose signup drop-offs in noisy feedback. For lean teams, post-launch maintenance and continuous improvement compete for the same time. Stitch Loop is positioned as a product agent that keeps working after ship.
You start by connecting GitHub and product context such as logs, user behavior, and feedback. Google says this can go through more than 180 integrations, custom MCP servers, or local CLI uploads. Grounded in that context, Stitch Loop runs evergreen loops toward measurable outcomes you care about. Three loops ship by default—cybersecurity, latency, and accessibility—and you can add custom ones, such as improving checkout conversion.
Each loop has four steps: Suggestions (evidence-backed opportunities), Solutions (options with tradeoffs and scoping, plus visual design where relevant), Build (code is written in an isolated cloud sandbox and opened as a GitHub pull request, with screenshots or videos when useful), and Reflect (it watches live product data, learns from positive and negative results, and keeps monitoring after a goal is met in case the metric slips). The post is explicit: “You always have the final call. Nothing merges until you review the PR.”
The security loop pairs Gemini 4 Carbon’s cyber security capabilities with CodeMender, Google DeepMind’s autonomous agent for improving code security. Together they scan the codebase, flag issues with evidence, and open a proposed fix for review. Stitch Loop can also act as a service layer for coding agents such as Google Antigravity, exposing persistent product context through the Stitch CLI and MCP. In Slack, @stitch can send daily insight digests and create new loops.
Usage
Stitch Loop is waitlist-only as of the announcement. Google has not published a general availability date or a price. Start with the official post Introducing Stitch Loop, then apply at stitch.google.com/loop/waitlist. Eligible Google AI Ultra users get priority, and Google invites feedback in the Google Labs Discord.
If you get access, the intended workflow is to connect GitHub and product data sources, keep or add loop goals, review suggestions and solutions, and decide on pull requests produced in the isolated sandbox. Officially, nothing merges automatically. If you already use a coding agent, you can pull Loop’s product context through the Stitch CLI or MCP rather than starting a separate process.
Treat the official docs and waitlist terms as the source of truth at the time you sign up. Stitch Loop is not a new foundation-model name; it is a long-running product-agent experiment under Google Labs’ Stitch lineup.