Google Labs has introduced Stitch Loop, an experimental artificial intelligence product agent designed to help businesses continuously improve software after it has been launched. Announced on November 18, 2026, Stitch Loop connects to a product’s codebase, live performance metrics, customer feedback and other data sources to identify problems and work on potential improvements around the clock.
Unlike conventional AI coding assistants that primarily help developers write code or complete individual programming tasks, Stitch Loop is designed to monitor what happens after software reaches production. It can identify security vulnerabilities, investigate performance problems, recommend improvements, write code in an isolated cloud environment and open GitHub pull requests for developers to review. Google says the system continues monitoring results after changes are made, allowing it to keep working towards predefined product goals.
Key Takeaways
- Google Labs announced Stitch Loop on November 18, 2026.
- It is an experimental, long-running AI product agent built around Google’s Stitch platform.
- The agent connects to GitHub, production metrics, user behaviour, logs and feedback.
- It supports more than 180 integrations, custom MCP servers and local CLI uploads.
- Three built-in automation loops cover cybersecurity, latency and accessibility.
- Developers can create custom loops for objectives such as improving checkout conversion.
- It writes code in an isolated cloud sandbox and creates GitHub pull requests.
- Human approval is required before changes are merged.
- A global waitlist is open, with eligible Google AI Ultra users receiving priority access.
What Is Google Stitch Loop?
Stitch Loop is a new Google Labs experiment that extends Google’s AI-powered Stitch design platform into continuous product development and optimisation.
Google Stitch helps designers, developers and founders turn ideas into user interfaces and product experiences. Stitch Loop addresses a different challenge: maintaining and improving those products after they become available to customers.
Software development does not end when an application launches. Businesses must continuously fix vulnerabilities, improve loading speeds, address accessibility problems and investigate why users abandon important processes.
For smaller companies, these responsibilities can consume significant engineering resources. Stitch Loop is designed to automate parts of this ongoing work by connecting AI agents to real product information and giving them measurable objectives.
Google describes it as a product agent that works continuously until the targeted outcomes are achieved.
How Does Stitch Loop Work?
Stitch Loop follows a four-stage process that combines product analysis, solution planning, software development and performance monitoring.
1. Suggestions: Identifying Problems
The first stage involves examining a product’s existing information to identify areas that require improvement.
After connecting its data sources, Stitch Loop can analyse information such as application logs, user behaviour, customer feedback and performance metrics.
For example, an online shopping platform might experience a decline in checkout completion rates. Stitch Loop could examine available product data and identify potential problems affecting the checkout experience.
Instead of relying only on a developer’s instructions, the agent uses evidence from the connected product environment to recommend possible improvements.
2. Solutions: Recommending Improvements
After identifying an issue, Stitch Loop proposes one or more potential solutions.
Google says these recommendations include information about trade-offs and project scope. Where relevant, the system can also provide visual design options.
This stage is important because a product problem can have multiple possible solutions.
For example, a slow checkout process might be caused by unnecessary interface elements, inefficient code or delays in an external service. Each problem requires a different response.
Stitch Loop is designed to help teams understand possible approaches before selecting a direction.
3. Build: Writing Code and Opening Pull Requests
Once developers select a solution, Stitch Loop writes the required code inside an isolated cloud sandbox.
It then creates a GitHub pull request containing the proposed changes for human review. Where relevant, the pull request can also include screenshots and videos showing the work completed by the agent.
This approach allows developers to inspect the changes before they become part of the production codebase.
Importantly, Stitch Loop does not automatically merge its pull requests. Google says the developer retains final approval authority.
4. Reflect: Monitoring Results Continuously
The final stage focuses on measuring whether the proposed changes actually improve the product.
After a change is made, Stitch Loop monitors relevant live product data and evaluates the results.
If the intended metric improves, the agent can use that information to refine its future recommendations. If the metric does not improve, it can investigate further.
The monitoring does not necessarily stop when a target is reached. Google says Stitch Loop continues watching the relevant metrics in case performance deteriorates again.
This feedback loop is what distinguishes the experiment from a conventional AI coding assistant.
Three Built-In AI Automation Loops
Google is launching Stitch Loop with three predefined automation loops. Teams can also create their own loops based on specific business objectives.
| Built-in loop | What it is designed to do |
|---|---|
| Cybersecurity | Identify code vulnerabilities and propose security fixes |
| Latency | Detect performance problems and improve application speed |
| Accessibility | Identify accessibility issues and recommend improvements |
| Custom loops | Target objectives such as improving checkout conversion |
Cybersecurity Loop Uses Gemini 4 Carbon and CodeMender
Security is one of the most important applications of Stitch Loop.
The system includes a dedicated cybersecurity loop that combines Gemini 4 Carbon’s cybersecurity capabilities with Google DeepMind’s CodeMender agent.
The tools work together to scan codebases for potential vulnerabilities, identify issues and prepare proposed fixes.
Stitch Loop can then open a GitHub pull request containing the suggested changes and supporting evidence.
For smaller development teams, this could help address a common challenge: maintaining software security while simultaneously developing new features.
However, identifying a vulnerability and proposing a fix does not automatically guarantee that the resulting application is secure. Developers will still need to review changes and conduct appropriate security testing.
Latency Loop Focuses on Application Performance
The latency loop is designed to help teams identify performance problems that affect how quickly applications respond.
Slow-loading pages, inefficient requests and delays in important user journeys can negatively affect customer experience.
By connecting to live performance metrics, Stitch Loop can look for areas where improvements may be possible and propose changes.
For businesses operating customer-facing applications, even relatively small performance improvements can matter when they affect large numbers of users.
Accessibility Loop Helps Improve User Experiences
The accessibility loop focuses on identifying problems that can make software difficult to use for people with different accessibility requirements.
These issues can include interface elements that are difficult to navigate, inadequate accessibility support and problems with how users interact with an application.
Automating parts of accessibility monitoring could help teams identify problems earlier instead of waiting for customer complaints or manual audits.
Stitch Loop Supports More Than 180 Integrations
One of the important aspects of Stitch Loop is its ability to connect with existing development and product-management systems.
Google says users can connect GitHub and product context through more than 180 integrations. The platform also supports custom Model Context Protocol (MCP) servers and local command-line interface uploads.
This means teams can bring information from different parts of their technology stack into a common product context.
| Integration capability | Purpose |
|---|---|
| GitHub | Connect codebases and manage pull requests |
| Product logs | Provide information about application behaviour |
| User behaviour data | Help identify problems in customer journeys |
| Customer feedback | Surface recurring complaints and improvement opportunities |
| Custom MCP servers | Connect additional tools and information sources |
| Local CLI uploads | Supply product context through local development workflows |
The availability of these connections could make Stitch Loop useful for teams that already rely on multiple development and analytics tools.
Instead of requiring businesses to replace their existing systems, Google is positioning the experiment as an additional layer that can work alongside existing workflows.
Google Stitch Loop Can Work With Antigravity and Slack
Google is also allowing Stitch Loop to fit into existing AI development workflows.
Integration With Google Antigravity
Stitch Loop can work as a service layer for coding agents such as Google Antigravity.
Through the Stitch CLI and MCP, coding agents can retrieve persistent product context from Stitch Loop.
This could help developers combine different AI models and coding tools while maintaining access to information about the product’s goals, performance and existing problems.
The approach also supports mixed-model workflows, where different agents can contribute to separate parts of a development process.
Slack Integration
Google is bringing Stitch Loop into Slack, allowing teams to interact with the system through an existing workplace communication platform.
Users can mention @stitch to receive daily insight digests and create new loops.
This could make it easier for product managers and engineering teams to monitor recommendations without constantly switching between different applications.
Stitch Loop vs Traditional AI Coding Agents
The biggest difference between Stitch Loop and conventional coding assistants is the stage of the software lifecycle they are designed to address.
Traditional coding assistants primarily help developers generate code, debug problems and implement requested features. Stitch Loop is designed to remain involved after software has been released.
| Capability | Traditional coding assistant | Stitch Loop |
|---|---|---|
| Generate code | Yes | Yes |
| Debug software | Yes | Designed to identify issues continuously |
| Monitor live product metrics | Not typically its central function | Yes |
| Analyse user behaviour | Depends on connected tools | Yes, through connected product context |
| Open GitHub pull requests | Supported by some tools | Yes |
| Monitor post-launch improvements | Not typically continuous | Yes |
| Built-in security, latency and accessibility loops | Depends on the tool | Yes |
| Human approval | Depends on workflow | Required before merging |
The distinction is significant because businesses often struggle with post-launch maintenance rather than simply writing new code.
Stitch Loop attempts to make continuous optimisation part of the AI-assisted development process.
Stitch Loop Availability: How to Join the Waitlist
Google has opened a global waitlist for Stitch Loop.
Eligible Google AI Ultra subscribers will receive priority access, while other users can register to express interest in the experiment.
Users can join through Google’s official waitlist:
The experiment is being introduced through Google Labs, which means its features, availability and capabilities may evolve as Google collects feedback.
Google has not announced a general public launch date or a complete commercial pricing structure for Stitch Loop.
What Stitch Loop Means for Startups and Small Development Teams
Stitch Loop could be particularly relevant for startups and smaller software companies that cannot maintain large engineering, security and quality-assurance teams.
A small team building a SaaS product, for example, may spend most of its time developing features and acquiring customers. Monitoring application performance, reviewing security issues and investigating customer complaints can become additional responsibilities.
An always-on product agent could help identify these problems and prepare potential solutions.
Lower Operational Workload
Continuous monitoring and automated recommendations could reduce the time developers spend manually investigating recurring problems.
Faster Product Improvements
The ability to connect product metrics with code changes could help teams move from identifying a problem to testing a potential solution more quickly.
More Accessible Software Maintenance
Smaller companies could gain access to automated security and accessibility assistance without building every capability internally.
However, these benefits will depend on how reliably Stitch Loop identifies problems, understands the connected data and proposes safe changes.
Risks and Limitations of Autonomous Product Agents
Despite its potential, Stitch Loop introduces important questions about the use of AI agents in production software environments.
Incorrect Recommendations
Product metrics can be misleading. A decline in conversions, for example, might be caused by seasonal demand, changes in advertising traffic or external service failures rather than a problem in the application’s code.
An AI agent must distinguish between correlation and causation before recommending changes.
Security and Data Access
Connecting an AI agent to source code, logs and customer behaviour data creates additional security and privacy considerations.
Businesses will need to carefully manage permissions, access controls and the information shared with external systems.
Human Review Remains Essential
Google’s decision to require human approval before pull requests are merged is an important safeguard.
An automated code change can introduce new bugs even when it is intended to fix an existing problem. Developers should review the proposed changes, run tests and verify the results before deployment.
Optimising the Wrong Metric
A product agent that focuses too heavily on a single metric could recommend changes that improve that number while damaging the broader user experience.
For example, increasing checkout completion should not come at the expense of customer trust, accessibility or security.
Businesses will need to define meaningful objectives and monitor more than one performance indicator.
The Bigger Picture
Stitch Loop reflects a broader shift in artificial intelligence from tools that generate content or code towards agents that work continuously towards measurable outcomes.
The traditional software development process separates building, launching, monitoring and maintaining a product. Google is attempting to connect these stages through an AI agent that can use production information to identify problems, propose changes and measure results.
This could become an important direction for AI-powered software development, particularly as businesses increasingly experiment with autonomous coding agents.
The launch also expands Google’s strategy around Stitch. Instead of focusing exclusively on turning ideas into interfaces, Google wants its platform to support a larger part of the product lifecycle, from initial design through continuous optimisation.
Looking Ahead
The next major test for Stitch Loop will be its ability to deliver reliable improvements in real-world software environments. Its performance will depend on the quality of connected data, the accuracy of its recommendations, the safety of generated code and its ability to distinguish genuine product problems from temporary changes in user behaviour or external systems.
For startups and development teams, the idea of an AI agent that continues monitoring and improving a product after launch could reduce some of the workload associated with software maintenance. However, human oversight will remain essential, especially for security-sensitive applications and changes affecting customers. Stitch Loop signals Google’s ambition to move AI coding beyond generating code towards continuously helping teams build and maintain better products.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



