Google is expanding its Stitch platform beyond designing and building user interfaces with a new experiment called Stitch Loop. Introduced on November 18, 2026, the Google Labs tool is a long-running AI product agent designed to monitor live software, identify opportunities for improvement, and continuously optimize applications after they have been shipped.
Unlike conventional coding agents that primarily help developers write new features and submit code for review, Stitch Loop focuses on what happens after a product goes live. It connects to a team’s codebase and real-world product data to identify bugs, address security vulnerabilities, improve performance, and uncover potential usability issues without requiring developers to manually monitor every aspect of their software.
How Stitch Loop works
Stitch Loop connects to GitHub and a product’s existing context, including application logs, user behavior, and customer feedback. Teams can bring this information into the platform through more than 180 integrations, custom Model Context Protocol (MCP) servers, or local CLI uploads.
Once connected, the agent runs continuous optimization loops aimed at measurable product goals. Google provides three default loops covering cybersecurity, latency, and accessibility, while teams can create additional custom loops based on their requirements.
Each loop follows a four-step process:
- Suggestions: Stitch Loop examines available product data and identifies opportunities to improve the software. Suggestions are grounded in evidence from the connected product rather than relying solely on generic recommendations.
- Solutions: For each opportunity, the agent proposes one or more possible approaches, outlining their trade-offs and scope. Where relevant, it also provides visual design options.
- Build: Once a team selects an approach, Stitch Loop writes the required code inside an isolated cloud sandbox and creates a GitHub pull request. Depending on the changes, the pull request can also include screenshots and videos showing the proposed work.
- Reflect: After the changes are reviewed and implemented, Stitch Loop monitors live product data to evaluate their impact. It uses positive and negative results to refine subsequent recommendations and continues monitoring even after the original goal has been achieved.
Importantly, Stitch Loop does not automatically merge its changes into a production codebase. Developers retain control over the process, and every pull request must be reviewed and approved before it can be merged.
Security, performance, and accessibility get dedicated AI agents
One of Stitch Loop’s most notable capabilities is its built-in security optimization loop. Google is combining the cybersecurity capabilities of Gemini 4 Carbon with CodeMender, Google DeepMind‘s autonomous code-security agent, to identify vulnerabilities in connected codebases.
When Stitch Loop detects a potential security issue, it provides supporting evidence and prepares a proposed fix through a GitHub pull request. This gives development teams another way to identify and address security weaknesses before they turn into incidents.
The other built-in loops focus on latency and accessibility, two areas that can directly affect the experience of people using a product. Teams can also create custom loops for specific business objectives, such as improving checkout conversion rates.
The idea is to move beyond treating software maintenance as a series of occasional development tasks. Instead, Stitch Loop can continuously examine how an application performs and identify changes that could make it better.
Works with existing coding agents and Slack
Google is also positioning Stitch Loop as a service layer that can work alongside existing development tools rather than requiring teams to replace their current workflows.
For example, developers can connect it with coding agents such as Google Antigravity. Through the Stitch CLI and MCP, other coding agents can access persistent product context from Stitch Loop, allowing teams to combine different AI models and agents in their development processes.
Slack integration is another part of the experience. Teams can use the @stitch command to receive daily insight digests and create new optimization loops directly within their existing collaboration environment.
This approach could make continuous product monitoring more accessible to smaller development teams that do not have dedicated specialists watching every metric around the clock.
Stitch Loop waitlist opens globally
Stitch Loop is currently an experimental Google Labs product. Its waitlist opened globally on November 18, 2026, with eligible Google AI Ultra subscribers receiving priority access.
Interested developers and product teams can join the waitlist through the official Stitch Loop website. Google is also collecting feedback through its Google Labs Discord community.
With Stitch Loop, Google is exploring a different role for AI in software development. Rather than stopping when an application is built and shipped, an AI agent can continue monitoring its performance, proposing improvements, and preparing code changes as a product evolves. Human developers still make the final decisions, but much of the ongoing work involved in keeping software healthy could eventually become continuous and increasingly automated.
Discover more from GadgetBond
Subscribe to get the latest posts sent to your email.
