Background coding agent
A coding agent that runs a task on its own, on managed machines, and returns a result you can review, usually a pull request.
A background coding agent works like hiring a contractor. You hand over a task. The work happens somewhere else. What comes back is a finished change to review. Editor assistants work the other way, with a human watching every step live.
The category went mainstream through 2025 and 2026, as GitHub, Cursor, OpenAI, Anthropic, and others added remote runs. The defining traits are simple. The run happens off your machine. It survives you closing your laptop. Its output arrives through the normal review channel, usually a PR with logs and diffs attached.
The economics differ from pairing. The human cost sits at the two ends: writing a clear task, then reviewing the result. That is why background agents reward clear specs and punish vague asks.
Related terms
See also: How a Syndai run works