Claude Code's New Agent View Puts Every Parallel Session on One Screen

Running multiple Claude Code sessions in parallel used to mean stacking terminal tabs, juggling tmux windows, and constantly switching context to check whether any agent had stalled waiting for a decision. The mental overhead grew with every additional session, and the failure mode was simple: an agent could sit blocked for minutes before you noticed. Anthropic's new Agent View, shipping in Claude Code 2.1.139 on May 11, 2026, addresses that directly by consolidating every active session into a single scrollable list with live state indicators, so developers can see at a glance which agents are running, which are waiting, and which need attention.

The feature ships as a research preview, and its arrival is squarely aimed at developers who are already using Claude Code the way Anthropic intended the tool to evolve: multiple agents working independently on separate tasks while a human supervises at the edges, stepping in only when the agent surfaces a question or hits a permission boundary. For those users, Agent View is a direct response to a workflow friction that had no clean solution before.

A Single List for Every Running Agent

Opening Agent View requires either running claude agents from the command line or pressing the left arrow while inside an active session. The resulting screen lists every Claude Code session currently running on the machine, along with each session's current state. The states cover the full lifecycle of a session: running, waiting, completed, failed, idle, and stopped. A session marked waiting is the signal that matters most — it means the agent has paused and needs a human response before it can continue, whether that is answering a clarifying question or approving a permission request. From the Agent View list, developers can reply inline without attaching to the full conversation, or attach to the session directly if the context warrants a more hands-on intervention.

Background sessions are launched with the claude --bg flag followed by a task description, which spins up a new agent process without occupying the current terminal. That agent runs independently and appears in the Agent View list with its status updating as it works. The combination of background launch and centralized status display creates a supervision model closer to a job queue than a conventional interactive terminal session. You dispatch tasks, let agents work, and check the list periodically rather than babysitting each window in sequence. The routines and desktop redesign update released the following week extended this pattern further with cloud-run automations that do not require the developer's machine to stay open at all.

Mixed Reception from Developers

Developer response has been positive on the concept but more divided on the execution. The core use case — supervising five or more agents simultaneously without losing track of which ones are blocked — is widely acknowledged as a real problem that Agent View solves. The list view, the state labels, and the inline reply option all work as described, and developers running large parallel workloads at the Code with Claude San Francisco conference earlier in May had already been asking for exactly this kind of session-level overview. For that audience, the feature landed well.

The criticism is coming from two directions. Developers working on smaller projects — one or two parallel sessions at most — find that switching to Agent View adds a layer of interface navigation that does not pay off at low session counts. The friction is low but not zero, and for those developers the old tab-switching approach remains faster. The larger concern is the absence of VS Code integration. A substantial portion of Claude Code users work inside the editor rather than a standalone terminal, and Agent View is currently a CLI-only feature. That gap has generated consistent feedback in developer forums and was the focus of a piece by The New Stack that asked why, despite the dashboard's clear utility, many developers were not yet convinced. Until Agent View surfaces inside the editor, it will reach a narrower slice of Claude Code's actual user base than Anthropic likely intends.

"For anything beyond two parallel sessions it's a genuine upgrade. But Anthropic needs to get this into VS Code before it reaches the majority of their users." Developer community feedback, May 2026

Part of a Larger Claude Code Overhaul

Agent View did not arrive in isolation. Claude Code 2.1 has been the most active development period for the tool since its initial release, with a sequence of features that collectively push the product toward a parallel-agent model as the default way of working. The new /goal command, released the same day as this article, lets developers set a high-level objective and have Claude Code pursue it autonomously across multiple sessions for hours without check-ins, surfacing only when it genuinely needs human input. Agent View is the supervisory layer that makes running several such goal-driven sessions at once tractable rather than chaotic.

The broader direction Anthropic is pursuing is a workflow where a developer specifies what needs to be done, dispatches agents to handle it, and monitors progress from a single interface rather than managing individual sessions manually. That vision was articulated explicitly at the Code with Claude SF conference in May, where Anthropic's engineering team described the parallel-agent model as the near-term target for how Claude Code should function at scale. Agent View, the /goal command, and the updated routines system are three components of that architecture arriving in close succession. Whether the VS Code gap gets closed quickly will determine how broadly that architecture lands beyond the subset of developers who live in a terminal.