Google Labs, a subsidiary of Google, launched the experimental agent Stitch Loop yesterday (November 18), which can connect code repositories with real-time online data to continuously identify issues, propose improvements, and write code after software deployment.

Unlike tools that terminate after completing a single task, this agent operates continuously around the clock based on the team’s objectives—even after the goal is achieved, it continues monitoring to prevent related metrics from declining again. Below is the accompanying demonstration video from IT Home:
Stitch Loop supports over 180 integrations and also allows data uploads via custom MCP servers or local command lines. Its workflow includes identifying improvement opportunities, suggesting alternatives, writing code, and tracking results, creating a continuous optimization cycle.

By default, Stitch Loop performs three cycles for network security, response latency, and accessibility. Users can also set additional cycles. Each cycle goes through the following four steps:
- Recommendation: Use evidence from the product to highlight opportunities for progress toward your goals.
- Solution: Each recommendation comes with one or more solutions, with code written in an isolated cloud sandbox and a GitHub pull request opened for your review. Where applicable, the PR will include screenshots and videos of the work completed by Loop.
- Reflection: Stitch Loop continuously monitors product data to help you achieve your goals and learn from every positive or negative outcome to improve insights in the next round. Once the goal is reached, it continues to monitor to prevent unintended changes.
- Execute: The agent will write code in an isolated cloud sandbox based on the user-selected plan, then submit a pull request to GitHub, optionally including screenshots or videos.
Google emphasizes that all changes must be reviewed by humans before being merged, and the final decision remains in human hands.

