Claude Code adds 'Agent View' and background tasks to power more autonomous agents

Anthropic added a new dashboard to Claude Code on May 11 that is designed to make longer-running, more autonomous coding sessions easier to manage. The company said the feature, called Agent View, lets developers see multiple sessions in one place, send work to the background, and respond when a session asks for input. (claude.com) Anthropic positioned the release as a way to reduce the clutter of terminal tabs and handoffs that come with running several coding agents at once. The launch extends a broader push Anthropic has been making since 2025 to support subagents, hooks, background tasks and other tools for longer-running agent workflows. (africa.businessinsider.com) ### What did Anthropic actually ship this week? Anthropic said on May 11 that Agent View is a new interface inside Claude Code that shows all active sessions in a single list. (claude.com) Each row displays the session, whether it needs user input, the latest response and the last interaction time, according to Anthropic’s product post. Developers can open it from the terminal with `claude agents`, peek into a session without attaching to it, and reply inline if the agent is waiting on a decision. The same update adds more explicit support for background work. Anthropic said users can move an existing session into the dashboard with `/bg` or launch a fresh background session directly from the command line with `claude --bg [task]`. (claude.com) Anthropic said the feature is available as a research preview on Pro, Max, Team, Enterprise and Claude API plans, with standard rate limits still applying. ### How does this fit into Claude Code’s longer push toward autonomous work? Anthropic had already been adding the pieces for longer-running agents before this week’s release. In a September 29, 2025 product update, the company said Claude Code gained checkpoints for autonomous operation, along with support for subagents, hooks and background tasks. (claude.com) Anthropic said those features were intended to let teams delegate more complex work while preserving control and the ability to rewind changes. Anthropic’s engineering team described the underlying problem in a November 26, 2025 post on long-running agents. The company said complex projects often span many context windows, and that agents can lose continuity between sessions unless they leave structured artifacts for the next run. (claude.com) Anthropic said its own harnesses rely on an initializer agent, a coding agent and context-management techniques such as compaction to keep work moving across hours or days. ### How far are users already taking these systems? Business Insider reported on May 13 that Boris Cherny, the lead engineer behind Claude Code, described an internal workflow that runs far beyond a single assistant session. (anthropic.com) Cherny said during a May 4 interview at Sequoia Capital that he typically has five to 10 sessions going and that “every night” he runs “a few thousand” AI sub-agents doing what he called “deeper work.” Cherny also said he monitors many of those sessions from his phone, according to the report. That account lines up with Anthropic’s own public framing at its May 6 “Code with Claude” event in San Francisco, where Cherny appeared in sessions focused on what was new in Claude Code and on live coding workflows. (anthropic.com) ### Does a better dashboard solve the trust problem developers keep raising? Anthropic’s own research suggests users are already granting the system more room to act. In a February 18, 2026 paper on agent autonomy, the company said the longest-running Claude Code sessions had nearly doubled in three months, from under 25 minutes to more than 45 minutes. (africa.businessinsider.com) Anthropic also said full auto-approve rises from roughly 20% of sessions among new users to more than 40% as users gain experience. The New Stack reported on May 13 that developers interviewed about Agent View saw the new interface as an observability improvement, not a complete answer to reliability and governance concerns. Anthropic reached a similar conclusion in its February paper, saying effective oversight will require new post-deployment monitoring infrastructure and new human-AI interaction patterns to manage autonomy and risk together. (africa.businessinsider.com) ### Where are developers supposed to learn these workflows? ZDNET reported on May 13 that Anthropic has been offering free Claude training courses, including material on Claude Code and subagents. The outlet said course lengths range from about 15 minutes to eight hours, and highlighted a 20-minute introduction to subagents as a practical starting point for developers trying to understand how to break work into parallel tasks. (anthropic.com) Anthropic also points developers to product documentation and event recordings as it expands the tooling. The company’s May 6 “Code with Claude” conference page says recordings are available, and Agent View’s launch post directs users to opt in through `claude agents` and then consult the docs. (thenewstack.io) Anthropic said Agent View is available now as a research preview, and the company’s public materials point users to the Claude Code docs and conference recordings for the next step. Boris Cherny, Dickson Tsai and other Anthropic speakers are already featured in those May 6 session archives as the company continues rolling out new agent features. (claude.com 1) (claude.com 2) (tech.yahoo.com)