Google Labs has introduced Stitch Loop, a long-running agent that connects to your codebase and live product metrics, finds what needs fixing, and keeps working on your product around the clock until your goals are met. A waitlist opens today for all users worldwide.
Google’s pitch starts from a gap. Coding agents help teams ship fast, but their job ends when a pull request merges. After launch comes patching vulnerabilities, speeding up checkout and figuring out why signups drop, and small teams struggle to keep up with that work.
How Stitch Loop works
You connect GitHub and your product context, such as logs, user behavior and feedback. Google lists more than 180 integrations, custom MCP servers and local CLI uploads. From that, the agent runs loops aimed at measurable outcomes. Three loops come built in, and you can add your own.
Every loop follows four steps:
- Suggestions: it finds opportunities backed by evidence from your product.
- Solutions: each suggestion comes with one or more options, with tradeoffs, scope and visual design choices where relevant.
- Build: after you pick a direction, it writes code in an isolated cloud sandbox and opens a GitHub pull request, with screenshots and videos when useful.
- Reflect: it tracks live data, learns from results and keeps watching after a goal is met in case the metric slips.
Google stresses that you keep the final call. Nothing merges until you review the pull request.

The built-in loops
The default loops cover cybersecurity, latency and accessibility. A custom loop could target something like checkout conversion.
The security loop pairs the cyber capabilities of Gemini 4 Carbon with CodeMender, a Google DeepMind agent that improves code security. Together they scan for vulnerabilities, flag each one with evidence and open a pull request with a proposed fix.
Fitting into existing tools
Stitch Loop can act as a service layer for other coding agents, such as Google Antigravity. Any agent can pull persistent product context through the Stitch CLI and MCP, which supports mixed-model setups.
In Slack, you can use @stitch to get daily insight digests and create new loops.
Where Stitch fits
Google says millions of designers, developers and founders already use Stitch to turn ideas into working interfaces. It describes Stitch Loop as a step toward agents that help lean teams from the first idea through continuous optimization.
For a small team, the appeal is obvious: someone has to watch metrics after launch, and an agent that proposes reviewed changes could take on part of that. The catch is review load. Every suggestion becomes a pull request someone must read, and quality depends on how good the agent’s evidence is. Google hasn’t published results from real users yet.
The problem Google says it is solving
Google argues that the hardest part of building a product starts after code ships. Teams must patch vulnerabilities, make checkout faster and work out why signups drop, with the evidence buried in noisy feedback. Lean teams, the company says, struggle to balance that maintenance against continuous improvement, even though that balance is what drives many of the most successful products.
Stitch Loop is built around the idea of an evergreen loop. Instead of a one-off task, the agent has a target metric, proposes changes, checks the result against live data and keeps going. After a goal is met, it keeps watching in case the number slips back.
The sandbox detail matters for trust. Code is written in an isolated cloud environment and arrives only as a pull request, so the change goes through the usual review.
How to get access
The Stitch Loop waitlist is open at stitch.google.com/loop/waitlist, and eligible Google AI Ultra users get priority. Google asks for feedback in the Google Labs Discord. Pricing and a general release date haven’t been announced, so the first signal will be how quickly Ultra users get in and what they report about the pull requests it writes.
Source: Google