Early look at Loop, new autonomous AI agent from Google

TestingCatalog
Read full story· 4 min
Google has introduced Stitch Loop, an autonomous AI agent that manages the entire product lifecycle from discovery to monitoring, continuously improving live products. It integrates with GitHub and various data sources to suggest, build, and test fixes before human approval. Stitch Loop leverages Gemini 4 Carbon and CodeMender to monitor security, latency, and accessibility, aiming to assist lean teams in post-launch product management.

What happened

Google has launched Stitch Loop, an experimental autonomous AI agent available through Google Labs. Stitch Loop is designed to manage and improve live products by handling tasks from product discovery to ongoing monitoring. The tool integrates with platforms like GitHub and various data sources to suggest, build, and test fixes, with human approval required before changes go live. Access is open globally via a waitlist, with priority for Google AI Ultra subscribers.

Key details

  • Stitch Loop uses Gemini 4 Carbon and CodeMender for monitoring security, latency, and accessibility.
  • It supports over 180 integrations, including GitHub, custom MCP servers, and CLI uploads.
  • Every workspace includes a default security loop that scans code and proposes fixes.
  • Teams can add custom loops for specific goals, such as improving checkout conversion.
  • Human review is required before any automated code changes are merged.

Why it matters

Lean product teams can now automate much of their post-launch management, reducing manual effort in monitoring and fixing live products. This offers these teams a way to maintain security, performance, and accessibility standards with fewer resources. The requirement for human approval before deployment helps organizations control changes while benefiting from continuous, AI-driven improvements.


More in Agents