Google Labs Launches Stitch Loop Product Agent With Default Security, Latency and Accessibility Loops

Google Labs introduced Stitch Loop, a new Stitch experiment that monitors production data and runs continuous loops until product goals are met. It includes default security, latency and accessibility loops built with Gemini 4 Carbon and CodeMender, and a global waitlist is open with priority access for eligible Google AI Ultra users.
Event version 1.
Key insights
Continuous product improvement agent
Stitch Loop is a product agent built for outcomes rather than just pull requests. It monitors production data and lets users build loops that work continuously until their goals are met.
Default security loop
Every Stitch Loop workspace includes a default security loop built with Gemini 4 Carbon and CodeMender. It scans the codebase, flags vulnerabilities with evidence, and opens pull requests with proposed fixes before they become incidents, with nothing merging until the user reviews the PR.
Latency and accessibility loops
A default latency loop watches live production data to spot bottlenecks, opens targeted PRs, and checks real metrics after deploys. A default accessibility loop watches for WCAG issues, listens to user reports, opens PRs with screenshots and videos of the fix, and verifies it holds in production.
Custom loops and integrations
Users can connect product data across more than 180 integrations and point a loop at outcomes such as checkout conversion or support tickets. Stitch Loop learns from every result and tailors suggestions over time.
Agent and workflow compatibility
Stitch Loop works as a service layer for coding agents such as Antigravity via the Stitch CLI and MCP, and can be brought into Slack for daily insight digests and task kickoff. The waitlist is open globally, with eligible Google AI Ultra users receiving priority access.