Google Stitch Loop: Always-On Agent Rewrites Code 24/7

Google Labs officially launched Stitch Loop on November 18, 2026, introducing a continuous agent that connects directly to codebases and live product metrics to identify issues, propose fixes, and write code around the clock. Unlike traditional development tools that complete a task and terminate, Stitch Loop operates indefinitely, monitoring deployed software to prevent performance regressions even after initial objectives are met.
How Stitch Loop Works
Stitch Loop integrates with over 180 data sources and supports custom uploads via Model Context Protocol (MCP) servers or local command lines. The agent follows a four-step optimization cycle: identifying improvement opportunities, suggesting alternative approaches, writing code implementations, and tracking results. By default, three parallel cycles run continuously, focusing on network security, response latency, and accessibility. Teams can configure additional cycles based on specific product needs.
Each cycle operates independently, allowing the agent to work on multiple optimization paths simultaneously. When Stitch Loop detects a metric decline or identifies a potential improvement, it generates a code proposal complete with context about the issue and expected impact. The agent then monitors whether the proposed change achieves the desired outcome, creating a feedback loop that informs future suggestions.
Human Review Requirements
Google emphasizes that all code changes generated by Stitch Loop require human approval before merging into production. The company positions the tool as an augmentation layer rather than a replacement for developer judgment. Engineers review proposed fixes through standard pull request workflows, maintaining full control over what enters the codebase. This approach mirrors emerging best practices in code review automation, where AI suggests and humans decide.
Integration With Antigravity
Stitch Loop functions as a service layer for Google's Antigravity platform, accessible through the Stitch CLI, MCP protocol, and Slack @-mentions. This architecture allows development teams to trigger optimization cycles from multiple interfaces without switching tools. The integration with Antigravity positions Stitch Loop as part of Google's broader developer tooling ecosystem rather than a standalone product.
Teams can configure Stitch Loop to monitor specific repositories, set priority metrics, and define thresholds that trigger new optimization cycles. The agent operates asynchronously, queuing suggestions and updates that developers can review during normal workflow rather than demanding immediate attention.
Availability and Access
Officially launched on November 18, 2026, Stitch Loop is now accepting global waitlist signups. AI Ultra subscribers receive priority access to the experimental tool. Google Labs has not announced general availability timelines or pricing beyond the initial waitlist phase. The launch follows a pattern of controlled rollouts for Google's AI development tools, allowing the company to gather feedback before broader distribution.
What This Means
Stitch Loop represents a shift from task-based coding assistants to persistent agents that operate as ongoing members of development teams. The always-on model addresses post-deployment monitoring gaps that traditional CI/CD pipelines miss, particularly around performance regressions and accessibility issues that emerge under real-world usage. By requiring human review for all changes, Google attempts to balance automation benefits with engineering oversight, a critical consideration as code agents take on more autonomous responsibilities in production environments.
on Emergent today





