Google Labs has introduced Stitch Loop, an experimental AI agent designed to monitor software performance continuously. The tool connects to codebases and real-time data to identify bugs, propose improvements, and generate code after a product launch.
Unlike task-specific tools, Stitch Loop operates around the clock based on team-defined goals. It maintains surveillance even after objectives are met to prevent performance regressions. The system currently supports over 180 integrations and allows data uploads via custom MCP servers or local command lines.
The agent follows a four-step cycle: identifying opportunities, proposing solutions, implementing code in isolated cloud sandboxes, and monitoring results. By default, it runs three cycles covering cybersecurity, response latency, and accessibility, though users can adjust the frequency.
For every proposed fix, Stitch Loop creates a GitHub pull request for human review. These requests may include screenshots and videos of the work performed. Google emphasizes that all changes require manual approval, ensuring that final decision-making power remains with the developers.


