AI-assisted coding has gone through three eras fast. Autocomplete, where you tab to accept suggestions. Synchronous agents, where you direct the model through a task in real-time. And now background agents: autonomous agents running in the cloud on sandboxed machines with full access to your dev, test, and deployment environment. You hand one a task, close your laptop, and come back to a PR.
This is shipping in real tools today. Codex spins up a cloud sandbox per task and hands back a PR with terminal logs. Claude Code on the web lets you kick off a task from your browser, close the tab, and come back later. Cursor spins up multiple cloud agents from the editor, each on its own branch while you keep working locally. Same idea across all of them: give the agent its own machine and let it work without you.
Why this matters
Synchronous agents are great but they occupy your attention, you're still watching them work, reviewing diffs, approving tool calls. Background agents break that 1:1 ratio. You can have 10 or 20 of them working while you think about architecture or just go to lunch. Last week I broke a feature into 6 tasks, dispatched them in parallel, and went for a walk. Came back to 6 PRs with passing CI. It feels more like managing a team than programming, and I'm still figuring out how I feel about that.
The part nobody talks about
What you tolerate individually becomes a disaster at scale. A flaky test that fails 5% of the time is annoying when you run it once, but when 20 agents are running tests simultaneously it blocks multiple runs per hour. All the tech debt you've lived with comfortably, the brittle fixtures, hardcoded paths, implicit dependencies, becomes a direct tax on agent productivity. Your infrastructure quality is your AI leverage ceiling, and nobody wants to hear that because fixing CI isn't glamorous.
Where this is going
A senior engineer ships maybe 2 meaningful PRs per day. With background agents you could be reviewing and merging 10-20, each from an agent that already ran tests before opening the PR. Some teams already have agents writing more code than humans, and within a year I think that'll be normal. Get your tests and CI in order first, because that's the foundation all of this runs on.