What happened
Google Labs introduced Stitch Loop on November 18, 2026. The announcement describes it as a long-running product agent that connects to a team’s codebase and live metrics, identifies what needs fixing, and keeps working toward goals after software has shipped. Google frames it as a complement to coding agents, whose work often ends when a pull request merges.
Teams begin by connecting GitHub and product context such as logs, user behavior, and feedback. Google says context can also come through more than 180 integrations, custom MCP servers, or local CLI uploads. Teams define measurable outcomes, then Stitch Loop runs persistent optimization loops. Three are included by default—cybersecurity, latency, and accessibility—and users can add custom loops, such as one focused on checkout conversion.
Each loop follows four stages. First, Stitch Loop surfaces improvement opportunities grounded in product evidence. It then offers one or more potential solutions, including tradeoffs and scope, and visual design options when relevant. After a person selects a direction, the agent writes code in an isolated cloud sandbox and opens a GitHub pull request. Google says a PR may include screenshots and videos of the work. The final stage is reflection: Loop monitors live product data as it pursues the goal and uses positive or negative results to refine later suggestions. When the goal is met, it continues watching in case the metric slips.
Google says teams retain control: no change merges until a person reviews the pull request. The security loop combines Gemini 4 Carbon’s cybersecurity capabilities with Google DeepMind’s CodeMender agent. It scans for vulnerabilities, flags findings with evidence, and opens proposed fixes as PRs. Stitch Loop can also act as a service layer for coding agents such as Google Antigravity. Teams can access persistent product context through Stitch CLI and MCP, or use Slack’s @stitch interface for daily insight digests and creating loops.
The announcement opens a global waitlist. Google AI Ultra users are eligible for priority access. Google does not announce a general-availability date or pricing in the announcement.
What the archive adds
New Runtime covered Stitch’s earlier evolution toward streaming design generation, targeted edits, and motion support on May 20, 2026. This is a new product direction rather than a restatement: Stitch Loop moves from interface creation toward continuous maintenance and optimization of deployed software.
Trend signal
Stitch Loop supports agent-ready software exposes capabilities: it connects an agent to explicit inputs—code, product metrics, integrations, and MCP servers—and routes its proposed work through reviewable pull requests. The announcement does not yet establish how reliably this boundary works in production, but the design makes permissions and approval part of the product workflow rather than relying only on prompts.
Why now
Google presents Loop as a response to the gap between shipping code and maintaining software in production. Its timing reflects a product-agent framing in which post-launch security, performance, accessibility, and measurable outcomes are treated as continuing work rather than a separate manual phase.
Connections to other events
Google’s earlier Stitch work made design context and editable interfaces central to its product-building workflow. Stitch Loop extends that direction: teams can use agent-built interfaces, then connect ongoing product context and operational goals to a separate loop that proposes changes after launch.
Why it matters
The announcement sketches a practical division of labor for product teams: an agent can continuously inspect signals and prepare scoped changes, while people retain the decision to merge. That could make maintenance work more visible and actionable for lean teams that struggle to keep security, latency, and accessibility improvements on the roadmap.
The important operational detail is the loop’s connection between evidence and code. Suggestions are grounded in product context, selected solutions become sandboxed changes, and subsequent metrics inform further work. Teams evaluating this approach should treat pull-request review and metric quality as essential controls: an agent can accelerate proposed fixes, but noisy signals or weak review can still produce bad changes. The experiment’s value will depend on whether those steps reduce the cost of reliable iteration without adding more review burden than they remove.
Limits and uncertainty
Google has announced a waitlist, not general availability, and has not stated pricing, service limits, or a launch schedule. The announcement provides no independent evaluation of vulnerability detection, latency improvements, accessibility outcomes, or the quality of proposed changes. It also does not quantify how the agent handles noisy or conflicting metrics, how teams configure permissions across integrations, or how much human review its pull requests require. The described workflow and capabilities remain company claims about an experiment.
Skill to build now
Write down one post-launch outcome your team can measure reliably, such as reducing a specific latency measure or improving checkout conversion. Identify the source of that signal and decide what evidence a reviewer would need in a pull request before approving a change. Those definitions will make any future optimization loop easier to constrain and assess.
Reader takeaway
Stitch Loop is Google Labs’ experiment in keeping an agent at work after deployment: connect product evidence, choose a goal, review sandboxed code changes, and let live results shape the next iteration. Its promise is not autonomous merging, but a persistent pipeline from operational signals to human-approved pull requests.