Newshot

Google launches Stitch Loop: a "product evolver" that never clocks off, watching your product and fixing bugs for you

AI summary

On November 18, Google Labs launched Stitch Loop, an experimental product agent that connects to code repositories and live online data to continuously spot problems, propose improvements and write code after software ships. Unlike tools that stop after a single task, it runs around the clock toward team-set goals and keeps monitoring after they are reached; it supports more than 180 integrations as well as custom MCP servers or local command-line uploads, and its workflow loops through surfacing opportunities, offering solutions (code written in an isolated cloud sandbox and opened as a GitHub pull request, with screenshots or videos where relevant) and reflecting on real-time data to sharpen the next round. By default it runs three loops covering network security, response latency and accessibility, with users able to set additional loops, and Google stresses that all changes must pass human review before merging, with the final decision remaining with people.

Translation(model translation, not the source's own text)

Google launches Stitch Loop: a "product evolver" that never clocks off, watching your product and fixing bugs for you

IT Home reported on November 19 that Google Labs yesterday (November 18) launched Stitch Loop, an experimental product agent that can connect to code repositories and live online data, continuously identify issues after software release, propose improvements, and write code.

Unlike tools that end after completing a single task, this agent runs around the clock around goals set by the team. After a goal is achieved, it continues monitoring to prevent the relevant metrics from declining again. IT Home includes the relevant demo video below:

Stitch Loop supports more than 180 integrations and can also upload materials through a custom MCP server or local command line. Its workflow includes discovering improvement opportunities, providing alternatives, writing code, and tracking results, forming a continuous optimization loop.

By default, Stitch Loop runs three cycles for cybersecurity, response latency, and accessibility (users can set additional cycle counts). Each cycle goes through the following four steps:

  • Suggestion: Stitch Loop presents opportunities to make progress toward the goal, based on evidence from the product.
  • Solutions: Each suggestion comes with one or more solutions, code is written in an isolated cloud sandbox, and a GitHub pull request is opened for your review. Where relevant, the PR will include screenshots and videos of the work completed by Loop.
  • Reflection: Stitch Loop monitors product data in real time, helps you achieve the goal, and learns from every positive or negative result to improve the next round of insights. Once the goal is achieved, it continues to observe to prevent mistakes in the changes.

In the execution phase, the agent writes code in an isolated cloud sandbox based on the solution selected by the user, then submits a code merge request to GitHub, attaching screenshots or videos when necessary. Google emphasizes that all changes must undergo human review before they can be merged, and the final decision remains in human hands.

Related articles

  • Google Pics upgrades Nano Banana 2.1 model: more refined image generation, more accurate image editing
  • The best AI assistant for App side hustles: Google launches a series of tools targeting the most headache

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by IT之家
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗