⚡️ On May 11 Anthropic released Agent View in Claude Code - a research preview that changes the very model of working with AI in development. Not an interface update - a shift in how you think about tasks in the first place
What is Agent View:
A single CLI dashboard that gathers all background agent sessions in one place. Background agents keep running regardless of the active tab - no context gets lost. Each row in the list shows: session ID, whether the agent is waiting for a reply, its last action and the time of the last interaction. An agent is stuck waiting for input - you see it right away. No more guessing "is it even still working?"
What you see for each session:
📌 Session ID - a unique identifier so you don't mix up tasks
📌 Waiting status - clearly shows if the agent is waiting for your reply right now
📌 Agent's last response - what it was doing, without opening every session
📌 Last activity time - when the last operation in this thread happened
Coordination between parallel agents isn't direct - they don't talk to each other. Instead: a shared task list, a common file on disk that all agents read and write to. Sounds simple, but it gives full transparency and an audit trail of every step, no magic
What a real pipeline looks like:
⚡️ Agent 1 - writes a new module, works in the background while you're busy with something else
⚡️ Agent 2 - hunts bugs in the already written code in parallel
⚡️ Agent 3 - digs through documentation or runs tests
⚡️ You - switch between sessions and pick up any of them right where it left off
A foreground agent blocks the main dialog until it's done - use it for operations where you need confirmation at every step. A background agent works autonomously and on its own declines actions that need permission you didn't grant in advance. For long-running tasks like PR monitoring - the next run time is shown right in the list, no need to guess whether the process is alive
Available on Pro, Max, Team, Enterprise plans and via the Claude API. Requires Claude Code v2.1.139+. Anthropic has already published a production case: a team of parallel agents built a C compiler from scratch. Not a concept - a real engineering project with a breakdown on anthropic.com/engineering
💻 Version: Claude Code v2.1.139+ - update right now
How I work as a dispatcher
I run 8 Claude Code agents every day, and I felt the dispatcher role before any convenient screen for it existed. I built coordination through Telegram: the marketer, copywriter and designer each live in their own group topic, and you can set a task by text or voice right in the right agent's topic. They don't talk to each other directly; one agent's result becomes another's input only through me or through a program
Above them sits my personal agent Jes. It has been running for 134 days and hands tasks to 7 other bots on its own, coming back to me only when there's a result or a choice to make. Since July 165 tasks have gone out this way. Anything risky, paid or public waits for my yes, so the dispatcher doesn't turn into a spectator watching agents break things
What I've learned
Parallel work only helps when each agent has its own role and its own area. Two agents in one file without rules give you conflict, not speed. So roles and limits come first, the number of agents second. Start with two whose areas definitely don't overlap, and add a third only when those two work without your corrections. The same logic works for clients: a clinic owner has 6 agents on his server, each with a clear area
💭 When Agent View shows up - it's not "open a few tabs". It's a change of role. You stop being a developer who writes code - you become a manager at HQ who sets tasks and controls execution. The difference is like between a driver and a fleet dispatcher. Anthropic is methodically building exactly this shift