feature
Agent view & background work
One screen for every Claude you have running — start jobs, send them to the background, and get pinged when one needs you.
Once you have more than one task going, you need a control tower. The agent view is a single screen showing every Claude Code session at once: what's running, what's finished, and what's blocked waiting on your answer. You can kick off a job, send it to the background, walk away, and jump back in only when it actually needs a decision — even from your phone.
A single conversation is easy to follow. Five of them at once is a juggling act —
and once you’re running subagents,
workflows, and a couple of long jobs, you genuinely have
five. The agent view (claude agents) is the control tower: one screen listing
every session, each tagged with what it’s doing — running, done, or blocked
waiting on you.
That last state is the useful one. Instead of watching a task crawl, you send it to the background and get on with something else; the view keeps it alive and tells you the moment it finishes or hits a question only you can answer. You attend to the work when it needs attention, not before.
It reaches past the terminal, too. With notifications on, a completed run or a pending decision can ping your phone — so a job you started before lunch can let you know it’s ready while you’re nowhere near your desk. The shape of the work changes from “sit and wait” to “kick it off and get called back.”
why it helps Run many tasks at once without losing track of any of them.
examples
claude agentsclaude --bg --exec 'pytest -x'Run the full test suite in the background and ping me when it's done.tips & best practices
- Send slow jobs (a big test run, a long migration) to the background and let the view tell you when they're ready.
- The view surfaces what's *blocked on you* — so you only jump in at the moments that actually need a human.
- Turn on notifications and a finished task or a question can reach your phone, not just your terminal.