Agent View lets developers run and manage parallel coding sessions in Claude Code

Claude Code is Anthropic’s terminal coding agent — the thing that reads your repo, edits files, runs tests, and handles git from a CLI. The problem was never just getting one agent to code. It was managing several of them at once without turning your terminal into a mess of tabs, worktrees, and half-forgotten sessions. On May 11, Anthropic added Agent View, a new Research Preview that gives Claude Code a single dashboard for parallel sessions. It landed in Claude Code v2.1.139, and the entry point is simple: run `claude agents`. ### What actually shipped? Agent View is basically a session manager inside Claude Code. Anthropic describes it as “a single list of every Claude Code session,” including ones that are still running, waiting on you, or already finished. That sounds small, but it changes the feel of the product from one conversational thread at a time to a control plane for many threads. (github.com) ### Why was this awkward before? Parallel work in Claude Code already existed in pieces. Anthropic’s docs describe “agent teams” with shared tasks, inter-agent messaging, and centralized management, but that setup is still experimental and disabled by default. Developers could also juggle worktrees, background jobs, and separate sessions by hand. The catch is that coordination lived mostly in your own head. (github.com) ### So what does Agent View fix? It gives those scattered sessions a home screen. Instead of remembering which terminal is reviewing a PR, which one is fixing tests, and which one is blocked waiting for approval, you get one list that surfaces state. That matters because the bottleneck in multi-agent coding is often not model intelligence — it’s operator overhead. If you can’t see the queue, you can’t really orchestrate it. (code.claude.com) ### Why does “running, blocked, or done” matter? Because those are the states that decide whether an agent system feels autonomous or annoying. A coding agent rarely fails in a dramatic way. More often, it stalls on a question, waits for permission, or quietly finishes while you’re looking somewhere else. Agent View turns that into something closer to an air-traffic board — not flashy, but exactly the layer you need when several jobs are moving at once. (github.com) The analogy does real work here: the hard part is not one plane flying, it’s sequencing many planes safely. ### Is this bigger than a UI tweak? Yes — because Anthropic has been pushing Claude Code toward longer-running autonomous work for months. In September 2025, it rolled out terminal upgrades, a VS Code extension, and checkpoints for more autonomous development. Earlier this year, Anthropic also showed what parallel Claude workflows can look like at full scale: 16 agents, nearly 2,000 sessions, and a Rust-based C compiler built for the Linux kernel. (github.com) Agent View looks like the practical product layer that sits on top of that direction. ### Who is this really for? Not the person asking Claude to rename a variable. This is for developers using Claude Code as a small team of workers — one agent on bug fixes, one on tests, one on review, maybe another on docs. It also fits the people building with the Claude Agent SDK, where the whole point is to create autonomous agents that can edit code, run commands, and execute workflows programmatically. (anthropic.com) ### What’s the limitation? Anthropic is calling Agent View a Research Preview, which usually means the workflow is useful but still settling. And the broader “agent teams” feature remains experimental. So this is not the final polished shape of multi-agent coding in Claude Code. It’s the first clear sign that Anthropic wants session orchestration — not just code generation — to be a core part of the product. (github.com) ### Bottom line Agent View matters because it solves a boring problem that becomes huge at scale: keeping track of parallel AI workers. Claude Code already had the ingredients for multi-agent development. Now it has a dashboard that makes those ingredients usable. (github.com)