Anthropic adds 'Agent View' to Claude Code CLI to manage parallel agents and workflows

Coding agents are starting to look less like chatbots and more like a small engineering team. That creates a new problem — once you can run several Claude Code sessions in parallel, the terminal gets messy fast. Anthropic’s answer is Agent View, a new Claude Code screen that pulls those sessions into one place and lets you manage them without hopping between windows. It landed on May 11 in Claude Code v2.1.139 as a research preview. ### What actually shipped? Agent View is a command-line dashboard for Claude Code. You open it with `claude agents`, or by pressing the left arrow from a session. The screen shows every background session, what each one is doing, which ones are waiting on you, what the last response said, and when you last interacted with it. ### Why did Claude Code need this? Because Anthropic has been steadily turning Claude Code into something that can work asynchronously. (github.com) Last year it added background execution, checkpoints, and a richer terminal interface so developers could hand off longer jobs instead of staying in a tight back-and-forth loop. Once you do that, the bottleneck shifts — not model capability, but human supervision. You need a control tower. (code.claude.com) ### What can you do from the new screen? The useful part is not just seeing sessions. You can peek into the last turn of a session, answer inline if the agent is blocked on a decision, and then let it keep running. If you need more context, you can attach directly to that session and read the full transcript. Basically, it’s a triage board for agent work. ### Why is “parallel agents” the real story? Anthropic has been telegraphing this shift for a while. (anthropic.com) Its Claude Code product page now says most code inside Anthropic is written by Claude Code, while engineers focus on architecture, product thinking, and “continuous orchestration,” including managing multiple agents in parallel. That language matters. It frames the developer less as the person typing every change and more as the person assigning, reviewing, and redirecting work. (claude.com) ### Is this just UI polish? Not really. The same v2.1.139 release also added a `/goal` command so Claude keeps working across turns until a completion condition is met. And the release notes mention agent-specific headers and observability fields for subagent requests. That tells you Agent View is sitting on top of a deeper multi-agent plumbing layer, not just a prettier list of tabs. ### How does this fit Anthropic’s broader strategy? (anthropic.com) Pretty neatly. Anthropic has spent the past year publishing about multi-agent research systems, long-running coding workflows, and agent teams that split work across many sessions. One internal experiment used 16 agents over nearly 2,000 Claude Code sessions to build a Rust-based C compiler that could compile the Linux kernel. Agent View is the product version of that worldview — less “talk to one model” and more “supervise a swarm.” (github.com) ### What’s the catch? Research preview means early, not finished. Anthropic is still figuring out how users understand and steer workflows that no longer live in one visible thread, and it has said subagents create new coordination and reliability problems. Agent View helps with visibility, but it does not solve the deeper challenge of making many agents dependable over long stretches of work. ### So what changes for developers? (anthropic.com) The immediate change is practical — fewer lost sessions, less context switching, faster supervision. But the bigger change is conceptual. If this model sticks, coding tools stop being single assistants and start behaving like managed teams. The developer’s job shifts upward — from writing every step to setting goals, checking progress, and stepping in only when judgment is needed. (code.claude.com) (anthropic.com)