DEV Community

IA Compa
IA Compa

Posted on Originally published at iacompa.com

Google Labs launches Stitch Loop, an always-on product agent that fixes security, latency and accessibility issues

Google Labs launched Stitch Loop on November 18, 2026, a long-running product agent that watches a team's codebase and live product metrics around the clock — and opens GitHub pull requests to fix issues it finds. Google opened a global waitlist today, with priority for Google AI Ultra subscribers.

What changed with Stitch Loop

Google Labs launched Stitch Loop on November 18, 2026, a long-running product agent that stays connected to a team's codebase and live product metrics, and opens GitHub pull requests to fix problems it detects.

Stitch Loop ships with built-in loops for security — powered by Gemini 4 Carbon and DeepMind's CodeMender — and for performance and accessibility. It uses the team's data and feedback from each loop to improve future work and adapt to changing codebases, product goals and requirements.

Loop owners can set goals, for example increasing conversion rates or removing performance bottlenecks. Stitch Loop then scans for issues that fit the goal, opens a pull request with the fix and a summary of the problem, and lets the owner accept or reject the change. Teams can also build custom loops with natural-language rules, and Stitch Loop can update them as priorities and codebases change.

Stitch Loop uses Stitch's design system — the same one used by the Google Labs Stitch app — so fixes are consistent with a team's existing interface. It is built on Gemini 4 Carbon, Google's new coding model, and on Stitch's earlier multimodal design models.

Why Stitch Loop matters for businesses

Why it matters: routine product maintenance — dependency and vulnerability fixes, slow pages, accessibility gaps — usually competes with feature work, and smaller teams often defer it until it becomes a problem. An always-on agent that fixes issues and opens pull requests for review could clear that backlog without adding headcount.

Put it to work:

Sources

Originally published at IA Compa.

Top comments (0)