AI NewsDev toolsAnnouncement

Google Labs launches Stitch Loop, an always-on product agent that reads live metrics

Google Labs launched Stitch Loop, an agent that reads a codebase and live metrics. It opens pull requests for a human to approve.

AI News

Editorial2 min read

LinkedInX
Google Labs launches Stitch Loop, an always-on product agent that reads live metrics

Why it mattersKeeping a feature healthy after release takes real time. A tool that reads the live metrics and writes the fix is a new way to spend less of it.

Releasing a feature is not where the work ends. Someone still has to read the live metrics. Someone still has to find the slow page. Someone still has to file the security patch. Google Labs says its new product agent does that part on its own. It only asks for a human on the pull request.

Google Labs launched Stitch Loop on 18 November 2026. Director of product management Justin Toupin wrote the announcement. He describes an always-on agent that connects to your codebase and your live product metrics. It picks issues to work on and writes the code in a sandbox. Then it opens a GitHub pull request for the team to approve.

What it does

Google describes four steps the agent repeats on its own. It reads the live product data and names an issue worth fixing. It proposes a solution with the trade-offs written out. It writes the code in an isolated sandbox and opens a pull request. Then it watches the live data to see whether the change worked, and starts again.

Three ready-made loops come with it. One scans for security problems. It is built on Gemini 4 Carbon and CodeMender. Google says CodeMender is the autonomous code security agent from Google DeepMind. A second loop looks for ways to cut latency. A third checks pages against accessibility rules. A team can also write its own loop.

How it fits into existing work

Google says Stitch Loop connects to more than 180 tools through a set of MCP servers. It talks to teams through Slack, using @stitch daily digests. It reads a repository through GitHub and files pull requests the same way. Every pull request waits for a human to approve it before it merges.

That is the safety line in Google's own words. The team keeps full control, because nothing goes live without a person clicking merge. The agent runs, the agent writes, and a person decides.

A team that watches a release this way spends less time on the parts a coding agent is good at. Reading logs, reading code, writing a small change: all of that moves off the person. More time is left for the parts the agent is not good at. The person decides what to fix next and whether a fix is worth the risk.

Stitch Loop is behind a waitlist at stitch.google.com/loop/waitlist, open worldwide. Google says Google AI Ultra users get priority. Google also runs a Discord channel for feedback from Labs users. Google has not said when the agent leaves the waitlist. The company has not said what Stitch Loop will cost, either on its own or inside a plan.

Source

Google Labs blog post by Justin Toupin

SourceGoogle

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project