If you’ve ever had six terminal windows open, each running a different AI agent, and found yourself frantically tabbing through them to see which one is waiting for your input, you know the pain. Claude Code’s new Agent View, released in version 2.1.139, aims to turn that chaos into a Kanban-like dashboard. One heavy user—who burned through 13.1 billion tokens, 606 sessions, and 38 projects in four months—calls it “the tmux for Claude Code.”
The feature is deceptively simple. It gives you a single panel that shows all active agent sessions, sorted into three columns: “Awaiting Input,” “Running,” and “Completed.” You can launch a new task in the background with claude --bg "task description" or /bg inside a session, switch between sessions with arrow keys and Enter, reply to a waiting session with Space, and kill one with Ctrl+X. Background sessions persist to disk, survive terminal closures, and are automatically placed in independent git worktrees—so no two agents accidentally step on each other’s code.
But what makes this more than a convenience feature is the problem it solves. It’s not about making Claude Code smarter; it’s about making the human less overwhelmed. In the old workflow, managing multiple agents meant relying on your own mental model: Which task is done? Which one hit a wall? Which one is silently waiting for my approval? You’d cycle through tabs, losing context and momentum each time. Agent View externalizes that overhead, turning it into a glanceable overview.
The real bottleneck in multi-agent coding isn’t the model—it’s the developer’s attention. This insight resonates beyond Claude Code. As AI agents proliferate in development pipelines—from Cursor to GitHub Copilot to custom AutoGPT setups—the skill that matters most shifts from “how to prompt” to “how to orchestrate.” Agent View is an early signal of that shift, treating the developer not as a prompt engineer but as a project manager of synthetic coworkers.
Of course, not everyone is thrilled. Some developers argue that managing six agents simultaneously is itself a bad practice, a form of premature parallelization. They point out that context-switching between tasks—even with a dashboard—still incurs cognitive load, and that spending energy on oversight might outweigh the benefits of parallelism. There’s truth in that. For simple, isolated workflows, a single agent is often faster and less error-prone. But for complex, multi-faceted projects—like building an app while reviewing a PR and researching a library—Agent View lets you run them all without losing your mind.
An understated detail is the git worktree integration. Each background session gets its own isolated working tree, meaning changes from task A won’t sneak into task B. This is a lifesaver when agents generate full file rewrites or refactoring. Without it, you’d risk conflicts or worse—silent corruption. Good tool design isn’t just about adding features; it’s about removing the cognitive tax of tracking invisible threads.
Looking ahead, Agent View hints at a future where developers don’t just talk to one AI, but supervise a team of them. Anthropic’s engineering lead Thariq described it as “a tmux for Claude Code,” but it’s more than that. Tmux manages terminal windows; Agent View manages agent state. It knows which tasks are stuck, which are waiting for permission, which are finished. As AI agents become more autonomous, the interface for human-agent teamwork will become as critical as the model itself.
If you’re already juggling multiple Claude Code sessions, this update is worth the download. If you’ve hesitated because you felt overwhelmed by parallel agents, Agent View lowers the barrier. The best productivity tools don’t make you work faster—they make you work smarter with less mental friction.
Try it: run claude agents or hit left arrow in any session. Your future self, the one currently hopping between six terminal tabs, will thank you.