📝Background Agents are AI agents that run a task asynchronously, without holding the user's terminal or chat, and report back when they finish or need input.
In an interactive session, the user waits while the agent works. A background agent is dispatched and left to run, whether in a separate process, a cloud sandbox, or a detached session. The user continues other work and reviews results later, typically as a summary, a branch, or a pull request.
Coding tools expose this in several forms. 📝Claude Code can run 📝Subagents and long shell commands in the background, dispatch cloud sessions from the web or mobile app, and track many sessions at once through 📝Claude Code Agent View. Comparable features exist in other coding agents.
Background execution changes how work is supervised. Because no one watches each step, background agents depend on clear task briefs, scoped permissions, and verification such as tests before their output is accepted. They are the building block for running agents in parallel and for hierarchies in which agents manage other agents.