Claude Code gets an agent view, and admits parallelism is the point

Anthropic shipped a small interface change that says a lot about where it thinks coding agents are heading. Agent view, announced today as a Research Preview in Claude Code, is a CLI screen for managing many Claude Code sessions in parallel: which ones are working, which are blocked on you, which are done.

Until now, running Claude Code agents in parallel has been a tmux-and-willpower exercise — multiple terminal tabs and a mental ledger of what each one is doing. The new view collapses that into a list where each row shows the session, its last response, when you last touched it, and whether it needs input.

What's actually new

According to Anthropic's post, you open it by pressing the left arrow from any session or running claude agents. From there:

• Peek and reply inline. Select a session to see its last turn; if it's waiting on a decision, you can answer without attaching. Press enter to drop into the full transcript.
• Background anything. Add an existing session with /bg, or skip the foreground entirely with claude --bg [task] to launch a fresh one straight into the background.
• Status at a glance. Rows show whether a session produced a PR, when a long-running agent is due to run again, and what's blocked on you.

Anthropic describes early use patterns as: dispatching several ideas at once and returning to a queue of PRs to review; running "PR babysitters" and dashboard updaters as looping jobs; and bouncing between a main task and quick side queries without losing place.

Why this matters more than it looks

The feature itself is modest — it's a list view. The signal underneath is that the assumed unit of work in Claude Code is no longer a session, it's a fleet. The product is being designed around developers who run many agents simultaneously and check on them the way you'd check a CI dashboard. That's a different mental model than pair-programming with one assistant, and it has implications for how rate limits, review workflows, and trust get structured downstream.

It also lands in a competitive context where every coding-agent vendor is trying to make parallelism legible without turning the terminal into mission control. A built-in dashboard, even a plain one, is a real ergonomic moat against the duct-taped tmux setups people have been posting for the last year.

The fine print

Agent view is available today on Pro, Max, Team, Enterprise, and Claude API plans as a Research Preview, opt-in via claude agents. Usual rate limits apply. Anthropic hasn't published a cap on concurrent sessions, and the post is light on what happens when you push the fleet hard — something worth watching as more users turn it on.

The interesting question now isn't whether developers want this. It's how many agents in parallel is actually useful before the bottleneck moves from terminal tabs to the human doing the reviewing.