Agent view in Claude Code

Overview

Agent view is a centralized interface inside the Claude Code CLI for managing multiple concurrent agent sessions. Released on May 11 as a Research Preview, it removes the need to track session state across separate terminal windows or IDE tabs — you can see which agents need input, are actively working, or have completed tasks from a single screen.

Key features

• See everything at once

  Running multiple Claude Code sessions previously meant switching between terminal windows or IDE tabs to check the state of each one. Agent view renders each session as a row showing its identifier, input status, last response contents, and interaction timestamp, so checking state and deciding what to do next happens in one place.

• Peek and reply without leaving

  You can peek at the last turn of any session without fully switching to it. Scan PR results or other long-running outputs by title to identify which agent needs your input, then reply inline to keep it moving.

• Background anything

  Send an active session to the background with the /bg slash command — it registers in Agent view and keeps running. To start a fresh background session without foreground interaction at all, run claude --bg [task] from the terminal.

• Dispatch multiple ideas in parallel

  Because Agent view absorbs the status-tracking burden, the number of sessions one developer can comfortably run at once goes up. The announcement highlights four usage patterns: scaling concurrent sessions with optional skills, managing long-running agents with scheduled task visibility, navigating between related sessions, and identifying completed work through status indicators.

claude agents is not the same as /agents

This is a common source of confusion — the docs explicitly note “Despite the similar name, this is separate from claude agents.”

Command	Where you run it	What it shows
claude agents	Terminal (CLI)	Agent view — status of background sessions
/agents	Inside an active session	Subagent panel — a Running tab (live subagents in the current session) and a Library tab (create and edit custom subagents)
/tasks	Inside an active session	Everything running in the background of the current session

The three commands target different things. Agent view operates at the session level, /agents operates on subagents inside a session, and /tasks tracks background work inside the current session.

How agent view fits among Claude Code’s parallel-work approaches

Claude Code offers five ways to run work in parallel, and agent view is one of them. The right one depends on who coordinates, whether the workers talk to each other, and whether they touch the same files.

Approach	Who coordinates	Worker-to-worker communication	File isolation	Use it when
Subagents	The main session (delegates and collects results)	Through the main session only	Optional worktree	A side task would flood your main conversation with search results, logs, or file contents
Agent view (claude agents)	You (dispatching to background)	None — sessions report only to you	Automatic worktree when editing files	You have independent tasks, want to hand them off and check status at a glance
Agent teams	A lead session (experimental, disabled by default)	Shared task list + direct messages	Not isolated — partition work yourself	You want Claude to plan, assign, and supervise a group of workers
Worktrees	You	None	Separate git checkouts	You’re running several sessions yourself, or subagents edit overlapping files
/batch	A planned split into 5–30 subagents — each opens a pull request	None	Worktree-isolated	A repo-wide migration or mechanical refactor describable in one instruction

Agent view’s distinctive behavior — automatic worktree integration: a dispatched session is automatically moved into its own worktree the moment it starts editing files, so file conflicts get quietly prevented.

These approaches combine: a session you dispatch via agent view can itself spawn subagents, which can each run in their own worktree. The five approaches above are not mutually exclusive.

Getting started

Prerequisite — Claude Code v2.1.139 or later. Check with claude --version. In earlier versions (e.g. v2.1.137), claude agents is a separate command that prints the configured subagent list and exits — the TUI does not appear. To upgrade, run claude install stable or claude install 2.1.139.

1. Run claude agents in your terminal to open Agent view.
2. From within an active Claude Code session, use the left arrow to switch between the session and Agent view.
3. Use /bg to move the current session to the background.
4. Use claude --bg [task] to launch a fresh background session.
5. See code.claude.com/docs/en/agent-view for the full reference.

Notes

• Research Preview — opt-in — Not turned on automatically. Run claude agents to access it. Behavior and UX may change before general availability.
• Plan availability — Pro, Max, Team, Enterprise, and Claude API. Standard rate limits apply with no extra purchase.
• Behavior on third-party providers (Bedrock, Vertex, Foundry, etc.) is not documented — the official agent view docs list supported plans as Pro, Max, Team, Enterprise, and Claude API only, with no statement about Amazon Bedrock, Google Vertex AI, Microsoft Foundry, or Claude Platform on AWS deployments. Users on those backends should verify in their own environment.
• More concurrent sessions = faster usage burn — Running many sessions in parallel accelerates token spend within your plan’s standard rate limits, so monitor remaining usage when you start dispatching background work.
• Separate from Remote Control — The previously announced Claude Code Remote Control is a delivery model for launching Claude Code sessions remotely. Agent view is an interface for managing multiple sessions inside your own terminal.