Google Labs opened a global waitlist for Stitch Loop on November 18, introducing an experimental product agent that works on software after launch. It connects a GitHub codebase with live product context, proposes improvements and builds changes for review. Eligible Google AI Ultra subscribers receive priority access.
Product data becomes the starting point for code changes
Google describes a workflow that begins with evidence from logs, user behavior and feedback. Teams can supply that context through more than 180 integrations, custom MCP servers or CLI uploads, then set measurable goals for the agent.
Built-in loops target security, latency and accessibility. Teams can also define their own objectives, such as improving checkout conversion. The security loop combines Gemini 4 Carbon with Google DeepMind’s CodeMender to identify code vulnerabilities and propose fixes.
TestingCatalog reports that coding agents can retrieve persistent product context through the Stitch CLI and MCP. Slack integration provides daily insight digests and a way to create loops. This puts the product’s observations within workflows a team already uses, including Google Antigravity6.4 out of 10.
A proposed fix still goes through a pull request
The agent first presents suggestions and possible solutions, including tradeoffs. After a team chooses a direction, it writes code in an isolated cloud sandbox and opens a GitHub pull request. Google says changes do not merge until the user reviews the PR.
The loop then watches live data to assess the result and continues monitoring after the goal is met. For a checkout-latency task, that design means the work includes measuring what happened after a change, rather than treating a completed patch as the outcome.
TestingCatalog’s Alexey Shabanov says he tested a pre-launch version on his publication’s site, where it suggested accessibility improvements and made small theme changes. That account offers an early usage example, rather than a controlled test of reliability across production systems.
Access begins with an experiment and a waitlist
The November 18 announcement opens registration worldwide; it does not establish general availability for every applicant. Google has not given a general-release date or a public price in the launch post.
For teams evaluating workflow automation, the relevant change is the proposed continuity between product evidence, code changes and subsequent measurement. Review quality and the usefulness of the selected goal remain central to whether that workflow improves a live product.