Google Stitch Loop: An Always-On Product Agent That Fixes Security, Speed and Accessibility via Pull Requests

Google Labs introduced Stitch Loop, an experimental agent that connects to a team's codebase and live metrics, runs security, latency and accessibility loops, and opens GitHub pull requests that humans must approve. A global waitlist is open.

Christopher Ort
By Christopher Ort · 3 min read
Share
A developer's hand on a keyboard in front of two monitors at night.
Image: AI-generated, i10X
On this page
  1. Quick Take
  2. In plain English
  3. What happened
  4. How it works
  5. Why it matters
  6. What's still open
  7. Sources
  8. i10x Perspective

Google wants an AI agent to keep working on your product after it ships. On 18 November 2026 Google Labs introduced Stitch Loop, an experimental, long-running product agent that connects to a team’s codebase and live metrics, finds what needs fixing and opens GitHub pull requests that a human must approve before anything merges.

i10X Quick Take

Quick Take

  1. What it is: an always-on product agent from Google Labs, part of a new direction for its Stitch design tool.
  2. Built-in loops: cybersecurity, latency and accessibility, plus custom loops such as improving checkout conversion.
  3. Security engine: pairs Gemini 4 Carbon’s cyber security capabilities with CodeMender, Google DeepMind’s code security agent.
  4. Human control: every change arrives as a pull request; nothing merges until someone reviews it.
  5. Access: a global waitlist opens today, with priority for eligible Google AI Ultra users.

In plain English

Most AI coding tools wait for you to ask for something. Stitch Loop watches your live app on its own, notices problems like slow pages, security holes or accessibility gaps, writes a fix in a safe test environment and sends it to you for a yes or no. It then checks whether the fix actually moved the numbers.

What happened

Justin Toupin, Director of Product Management at Google Labs, announced Stitch Loop in a Google blog post. “Millions of designers, developers, and founders already use Stitch to turn ideas into working interfaces in minutes,” he wrote, calling Stitch and Stitch Loop “steps toward our broader ambition for Stitch: creating agents to help lean teams tackle the hardest parts of creating great products.”

How it works

Teams connect GitHub and product context such as logs, user behaviour and feedback through more than 180 integrations, custom MCP servers or local CLI uploads. Each loop then runs four steps:

  • Suggestions: opportunities toward a goal, grounded in evidence from the product.
  • Solutions: one or more options with trade-offs and scoping, plus visual design options where relevant.
  • Build: code written in an isolated cloud sandbox and a pull request opened, sometimes with screenshots and videos.
  • Reflect: it monitors live data, learns from results and keeps watching after a goal is met in case the metric slips.

Stitch Loop can also act as a service layer for coding agents such as Google Antigravity: any coding agent can pull product context through the Stitch CLI and MCP. In Slack, @stitch posts daily insight digests and creates new loops.

Why it matters

For lean teams, post-launch work such as security fixes and performance tuning often gets dropped. A loop that proposes evidence-based fixes could cover some of that gap.

For developers, it shifts AI from writing new features to maintaining live systems, while keeping review in the familiar pull-request flow.

For Google, it is a showcase for Gemini 4 Carbon and CodeMender in a product setting, competing with third-party coding agents.

What’s still open

  • Pricing and usage limits after the waitlist phase.
  • How well the security loop performs on real codebases and how many false positives it creates.
  • What data access teams must grant and how product data is handled.

Sources

  • Google, “Introducing Stitch Loop: A new way to build and evolve products,” blog post by Justin Toupin, 18 November 2026

i10x Perspective

Coding agents are moving from the editor into operations. Stitch Loop’s bet is that an agent tied to live metrics, not prompts, will find the work worth doing. The pull-request gate is the right default; the open question is how many PRs a small team can actually review. More from Google.

Christopher Ort

Christopher Ort

Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU). He is the author of “The Style Penalty”, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.

Timeline

What happened before

  1. Release

    Gemini 4 Argon

    1.05M context · $2 / $10 per 1M tokens

    Google's most capable model for complex coding, agentic, math and science tasks.

    ai.google.dev (source, opens in a new tab)
  2. Release

    Gemini 3.8 Flash

    1.05M context · $0.75 / $3.75 per 1M tokens

    ai.google.dev (source, opens in a new tab)
  3. Release

    Gemini 3.7 Flash

    1.05M context · $0.75 / $3.75 per 1M tokens · Deprecated 8 Oct 2026 (API routes to Gemini 3.8 Flash)

    ai.google.dev (source, opens in a new tab)
  4. Release

    Gemini 3.6 Flash & 3.5 Flash-Lite

    1.05M context · $0.75 / $3.75 per 1M tokens

    ai.google.dev (source, opens in a new tab)
  5. Partnership

    Google to pay SpaceX $920M a month for Colossus compute (~$30B through 2029)

    With Google · Compute · $30B

    Per SpaceX's SEC filing: about 110,000 NVIDIA GPUs from October 2026 to June 2029, ramping through September at a reduced fee; either side may terminate after Dec 31, 2026 with 90 days' notice (CNBC, Bloomberg).

    cnbc.com (source, opens in a new tab)
Full Google timeline ›

Editor's picks

  1. GAO Report: ChatGPT, Claude, Gemini and Grok Produced Viable Attack Plans After Safeguards Were Bypassed
  2. ResolveAI SRE-1: A 27B Model Trained to Find the Root Cause of Production Incidents
  3. Reson8 Rumble-2-Flash: Open-Weight 0.5B Speech Recognition Model Covers 40+ Languages
  4. Pika Enact Lets Creators Direct AI Video Actors and Cameras in Real Time
  5. Google Stitch Loop: An Always-On Product Agent That Fixes Security, Speed and Accessibility via Pull Requests