WTF is Asynchronous Coding Agent?
A fire-and-forget coding agent that works in its own hosted environment and comes back with a pull request.
Also called: async coding agent, background agent, cloud coding agent
TL;DR
You describe a task, close the tab, and later a pull request appears. Unlike an agent in your editor, it runs on a server, clones the repo into a sandbox, iterates until tests pass, and hands you the diff to review. You can run several in parallel like a small team.
How it works
Defining traits: a hosted, sandboxed checkout per task; no human in the loop between kickoff and PR; output as a reviewable pull request. Because the agent cannot ask you questions mid-run, success depends on a clear task description, a repo with runnable tests, and a setup script so the environment builds. Security characteristics differ from local agents: credentials live in the sandbox, and network access is usually restricted.
Origin
- Coined / introduced by
- Google (Jules announcement); popularised by Simon Willison
- When
- May 2025
- Story
- Google called Jules an 'asynchronous coding agent'; Willison adopted the phrase as the category name.