cd /news/artificial-intelligence/google-builds-an-ai-that-never-stops… · home › topics › artificial-intelligence › article
[ARTICLE · art-172213] src=dev.to ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

Google Builds An AI That Never Stops Fixing Your App After Launch

Google Labs unveiled Stitch Loop, an experimental AI agent that monitors a live app's metrics after launch, picks a priority, tests a fix in a sandbox, and opens a pull request for human approval before checking whether the metric improved. The tool runs on Google's Gemini 4 Carbon model and CodeMender, and integrates with coding agents like Antigravity via the Stitch command-line tool and MCP. A worldwide waitlist is open with Google AI Ultra subscribers first, though Google has not announced pricing or a wider access date.

key takeaway

Stitch Loop ties post-launch fixes to live metrics and gates every change behind a sandboxed test and a human-approved pull request, and exposes product context and suggested fixes to other coding agents through the Stitch CLI and MCP.

read3 min views4 publishedNov 19, 2026

Google wants an AI that does not clock off when the code is done. Its new Google Labs experiment, Stitch Loop, keeps watching an app after it goes live, works out what is hurting it, tests a fix on its own and then asks a human to approve the change.

"Today's coding agents stop when code ships," Google said on X when it unveiled the tool on Nov. 18. "Stitch Loop, a new experiment from @GoogleLabs, keeps working after launch." Its product page sums up the pitch in one line: "Find what moves the needle. Loop until it does."

Think of it as a product manager and a junior engineer rolled into one bot. It turns scattered signals, such as user numbers and other live metrics, into one clear priority, ships a fix for it and then keeps checking whether that number moved.

Google says every change is tested in a sandbox first, a walled-off copy where nothing can break, and arrives as a pull request, a proposed code change that a person must sign off before it goes live. The work is always tied to goals the team sets.

Step What Stitch Loop Does
1. Watch Tracks live metrics after the app launches
2. Pick Turns scattered signals into one clear priority
3. Test Tries a fix in a sandbox where nothing live can break
4. Ask Opens a pull request for a person to approve
5. Check Monitors whether the metric actually improved

Also Read: Google Researchers Publish Method to Stop AI Agents Memorizing Their Own Tests Stitch Loop runs on Gemini 4 Carbon, one of Google's newest Gemini models, and CodeMender, Google's AI tool for finding and fixing security flaws in code.

It is also meant to work alongside other bots rather than replace them. Coding agents such as Google's Antigravity can pull product context and suggested fixes from Stitch Loop through the Stitch command-line tool and MCP, a common way for AI tools to talk to each other, "so your agent knows what to autonomously fix next," Google said. Teams can also get daily digests and start tasks from Slack.

This is the next step in the AI coding race. The first wave of agents wrote code when asked. Stitch Loop tries to own the job after that, deciding what to work on based on how real users behave. Google is aiming it at "lean teams," and its Stitch account is recruiting startups at the Seed or Series A stage with a live product as design partners.

The approval step is the safety valve. Letting an AI push changes to a live product without a human check would be a big risk, and Google has kept a person in charge of every change, at least for now.

The waitlist is open worldwide, with Google AI Ultra subscribers first in line, but Google has not given a date for wider access or a price. Watch how much freedom early users give it. If startups start letting it ship fixes with lighter checks, the line between a coding tool and an AI teammate that runs part of the business will get very thin.

Also watch rivals. Every big AI lab is selling coding agents, and Google is betting the winner will be the one that keeps working after launch, not just the one that writes code fastest.

**Read Next:** [Google Treats Its Own AI Agents Like Rogue Staff After Catching Cheats](https://www.fathom.news/ai-agent-safety-google/)

*Originally published on [Fathom News](https://www.fathom.news/stitch-loop/).*
── more in #artificial-intelligence 4 stories · sorted by recency
── more on @google 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/google-builds-an-ai-…] indexed:0 read:3min 2026-11-19 · —