Introducing background agents
Kick off a task and keep coding. Your agent works in the background and pings you when the diff is ready.
Today we’re shipping background agents in Arshas 1.8: hand off a task, keep working in the same repository, and review a finished diff when the agent is done. No blocked editor, no watching a spinner.
Why backgrounding matters
The first generation of coding agents had a dirty secret: they turned your editor into a viewing gallery. The agent worked, you watched. For a thirty-second edit that’s fine; for a forty-file dependency upgrade it means your most productive tool is occupied by your least interactive work.
Background agents break that coupling. Each run operates on an isolated snapshot of your repository, so the agent’s edits and yours never collide — even in the same files. Start a run, open a new task, or just keep coding.
How it works
Choose “Run in background” in the agent panel and the run moves to the runs sidebar, where you can watch progress — current step, files touched, test results — or ignore it entirely until the notification fires.
When the run finishes, Arshas rebases the agent’s diff onto your current working state. Anything you changed in the meantime is reconciled automatically, and true conflicts are flagged explicitly in the review so nothing merges blind.
What people use it for
In the beta, three patterns dominated: mechanical refactors (renames, API migrations), test backfills for legacy modules, and the “end of day queue” — kicking off a tedious task at five and reviewing a finished diff with morning coffee. Multiple background agents can run at once, so the queue can be more than one item deep.
Background agents are available today on every plan, including the free tier. Update to 1.8 and press ⌘I.