Claude Code Projects - stop managing sessions, talk to one agent

Canonical version: Claude Code Projects - stop managing sessions, talk to one agent.

If you use Claude Code seriously, you probably run several sessions in parallel. And you probably know the problem: YOU are the orchestrator. You open five sessions, try to remember which one does what, and copy context from one to the other by hand.

On 17 September 2026, Anthropic launched Projects in beta to fix exactly that. In this piece, I'll explain what it is, how it compares to the other multi-agent features of Claude Code, and what I'd keep an eye on.

What it is

A project is one conversation with Claude. You describe what needs doing, and Claude splits the work into threads by itself. Each thread runs as a parallel cloud session, context flows between them through the orchestrator, and the whole thing keeps going after you close your laptop. You can also ask it to be proactive or to run things on a schedule, and it mixes expensive and cheaper models depending on your preferences.

For now, it's available on desktop and web for select Pro and Max users. Anthropic says everyone gets it soon.

(Don't confuse it with the Projects feature of the Claude app, which is a shared knowledge base for chats. Same name, different thing 😅)

Thariq Shihipar, from the Claude Code team, explains that it brings the architecture of Claude Tag to Claude Code: one long-lived agent that holds the memory and delegates, instead of a pile of independent sessions.

Why I think it matters

Boris Cherny summarized it well: he stopped managing sessions. He sends thoughts as they come, and Claude splits them into threads. The project remembers how he works.

That's the real shift. You go from managing agents to talking to ONE agent that manages the others.

Ethan Mollick's experiment shows how far this can go. He asked a project (running Claude Fable 5.1) to pick famous historical mysteries and try to solve them. It launched research agents, chose the mysteries, spun up eighteen threads (one per mystery), and each thread launched its own agents for simulations and code breaking. A final set of skeptical agents fact-checked the write-ups. About a day of work, fully coordinated by the orchestrator. In his words, it basically creates an organization to solve your problem.

How it compares

Claude Code now has quite a few multi-agent features, so here's how I see them:

  • Claude Code Agent View: one screen to dispatch and watch background sessions. Useful, but you're still the orchestrator
  • Claude Code Agent Teams: local, experimental coordination between Claude Code instances
  • Claude Code Routines: saved cloud sessions with triggers. One task, not a living project
  • Dynamic workflows (ultracode): script-driven fan-out inside a single session, deterministic and repeatable
  • Projects: a persistent conversation that plans, delegates and remembers across threads

If you like control and repeatability, workflows are still the better fit. If you want to throw a messy goal at Claude and come back later, Projects is the one to try.

What I'd watch out for

  • Cost visibility. A project can fan out to dozens of agents. On a subscription, that can eat your limits quickly. On the API, it's your bill
  • Auditability. When eighteen threads each launch their own agents, can you still check what each one did, and why? That matters as soon as the work is more serious than a fun experiment
  • Beta is beta. Limited access, and the behavior will change

I'm really curious to test it on my own projects. My plan: start with a well-scoped goal (e.g., a documentation overhaul across one repository), check the threads it creates, and see how much I need to steer.

More details in my Claude Code Projects note.

That's it for today! ✨

References


About Sébastien

Ready to get to the next level?

Found this valuable? Share it with someone who needs it.

Join 6,000+ readers. Get practical systems for knowledge & AI. Free.

Subscribe ✨

Free: Knowledge System Checklist

A clear roadmap to building your own knowledge system. Subscribe and get it straight to your inbox.

6,000+ readers. No spam. Unsubscribe anytime.

Subscribe