TellWorldAI World Archive

Home›Agents

Agent

Stitch Loop

Stitch Loop is a long-running product agent from Google Labs. It connects to a codebase and live metrics and keeps improving software after launch, working until its goals are met.

Era Generative AI for everyone

Release date
Developer
Google

What it is

Stitch Loop is a long-running product agent from Google Labs. It connects to a codebase and to live metrics, finds what needs fixing, and keeps working on a product 24/7 until its goals are met.[1]

The idea behind it

Google's announcement argues that a coding agent's job ends when a pull request merges. Stitch Loop is positioned as work that continues after the software ships.[1]

Origin

Google Labs introduced Stitch Loop on 18 November 2026 as a new experiment and as part of a new direction for Stitch.[1] The announcement was written by Justin Toupin, Director of Product Management at Google Labs.[1]

Status

Stitch Loop is an experiment from Google Labs rather than a general release.[1]

Terms used in this entry

A coding agent is an AI model that writes, edits and tests code on a developer's behalf. Its usual unit of work is a pull request, which is a proposed set of code changes that a team reviews and merges into the main codebase. Once a pull request is merged, the code is considered delivered, and the question of what happens to the product next is usually handled by people watching the product.

Live metrics are measurements of a product while it is in use. Examples include error rates, how long pages take to load, and how often people finish a task they started. A system that connects to these metrics can notice when something is getting worse, compare it with earlier numbers, and decide what to change next. This differs from a one-off coding task, which ends once the requested change is made.

A long-running agent is one that keeps working across many cycles instead of answering a single request. Each cycle might look for a problem, propose a fix, check the result against the metrics and then continue. The idea of working until goals are met means the agent needs a clear target to measure against, so that it knows when a cycle has produced enough improvement and when to keep going.

Work that continues after a product ships is a different kind of task from building it. Software in real use shows problems that testing did not catch, and user needs change. Keeping a product in good shape after launch is therefore a steady job rather than a single milestone, and a tool built for that job has to watch the product, not just the code.

CapabilitiesCodeAgents & tool use

Ecosystem & regionDeveloper tools

20261

  1. ReleaseGoogle Labs introduces Stitch Loop

    On 18 November 2026 Google Labs introduced Stitch Loop, a long-running product agent that connects to a codebase and live metrics. It is an experiment rather than a general release.

Developed by1

Google

Notes and sources

  1. Introducing Stitch Loophttps://blog.google/innovation-and-ai/models-and-research/google-labs/stitch-loop/

Cite this page: Stitch Loop. TellWorld, https://www.tellworld.net/agents/stitch-loop. Updated 2026-11-20