Claude Code new release Agent View: a screen that manages all AI agents, ending multi-terminal chaos

Anthropic officially launched Claude Code Agent View on the 11th, a unified CLI console that allows developers to launch, monitor, and respond to multiple AI agent sessions simultaneously within a single interface.
(Background: Claude Code learns as it goes: this free website uses 11 lessons to teach you how to get started, no installation needed)
(Additional context: Top AI models are diverging: ChatGPT to C, Claude to B)

Managing more than five concurrent AI agents might have previously meant five terminal tabs and a task tracking system supported by memory. Anthropic’s Agent View feature, launched on the 11th, aims to consolidate this chaotic mental ledger into a single CLI screen.

After all, when the number of AI agents expands from one to ten, twenty, cognitive load becomes the bottleneck, not computing power itself.

One screen to replace tmux layouts

Before Agent View was introduced, Claude Code’s multi-agent workflow relied on developers organizing themselves: multiple terminal windows running different tasks, manual switching, remembering each window’s progress. This approach becomes unmanageable when the number of tasks exceeds three or four.

The core logic of Agent View is to bring all sessions into a visible list. Developers can press the left arrow key on any session, or run claude agents directly in the terminal, to open this control panel.

Each row in the list displays four pieces of information: the session’s name, whether it’s waiting for user input, a summary of the last response, and the last interaction time.

The “Waiting for Input” column is also crucial. In the past, when an agent was stuck waiting for a decision, developers wouldn’t know which task had stalled unless they manually checked each terminal periodically. Agent View surfaces this information prominently, turning multi-agent scheduling into an experience similar to managing a task kanban board.

Background execution enables agents to operate independently

Another key design feature of Agent View is its background execution mechanism. Developers can use the /bg command on any existing session to push it into the background; or start a new task directly with claude --bg [task description], skipping the foreground interface, allowing the agent to run without occupying attention.

The business logic behind this design is quite clear: the longer and more complex an AI agent’s execution time, the more developers need a “don’t watch it” mechanism. Early users’ patterns confirm this, with four high-frequency scenarios summarized:

First, mass parallel dispatch. Launch multiple tasks simultaneously, each with different skills, then return to the list to browse the ready pull requests.

Second, long-term task management. For monitoring PRs, dashboard auto-updates, and other cyclical work, Agent View directly shows the next execution time for these agents.

Third, quick switching between sessions. During ongoing tasks, press the left key to quickly open a subsidiary task or code query; after completion, right-click to return to the main thread. The Peek feature makes responses instantly visible upon arrival, without switching focus.

Fourth, tracking delivered results. The status indicators on the list, combined with Peek’s title previews, allow developers to quickly scan which sessions have produced PRs.

These four modes all point to the same need: as AI agents evolve from auxiliary tools to semi-autonomous executors, developers shift from “coding” to “scheduling agents,” which requires a dashboard rather than a row of terminals.

Enterprise scale is already underway

Currently, Agent View is available as a Research Preview, compatible with Pro, Max, Team, Enterprise, and Claude API paid plans, with standard rate limits applied. Official technical documentation has also been published.

The number of agents is no longer the issue; managing them is. Agent View is shifting this management cost away from developers’ mental accounts.