Claude Projects was expanded on October 9, 2026, when Anthropic's developer account ClaudeDevs announced that everyone on the waitlist with a Pro or Max plan has now been admitted, with the beta continuing to roll out to those two tiers. Its shape is a single ongoing conversation: you keep sending work to one place, and Claude splits it into threads that run in parallel and report back.
One coordinating conversation, many working threads
Without a project, coordinating several parallel sessions is your job: deciding what each session works on, repeating the same background at the start of each, and checking which finished or needs an answer. Projects hand that coordination to Claude itself. The project conversation receives tasks, decides whether to start a new thread or hand the work to a thread already active in that area, and tracks what threads report back. Each thread is a separate session with its own context window. Most are cloud sessions that work on their own branch and open a pull request when the work calls for one; when a task needs something only your computer has, that thread can run on your machine instead. Threads proceed in parallel, you can check in and steer them from your phone, and cloud threads keep going after you close your laptop.
Set context once, watch one overview
Project instructions and memory reach every new thread: which branch to target, how a thread should check its work, and what needs your go-ahead — stated once instead of pasted into every session. When you return, the Overview pane lines up every thread's state: which finished, which pull requests are ready for review, and which thread is waiting on you, alongside a Library tab for the files you added and the files threads produced. The intended uses are spelled out: one goal spanning many repositories, a steady stream of incoming bugs and review requests, or a build or migration too big for one session. Non-code work fits too — upload contracts or a support-ticket export, and threads deliver their write-ups back as files.
Count the usage cost before adopting it
The boundaries are equally clear. A single task that fits in one session is simpler as a plain cloud session; work where every task needs your machine belongs in local sessions; a scheduled one-off job does not need the project layer at all. On cost, projects draw on the same plan limits as other Claude Code sessions and use them faster — the documentation notes that a new project runs every thread on Opus at high effort by default, the fastest way to burn through a plan. The sensible start is to check the thread model and effort in project settings, ask Claude to propose threads before starting them, run only a few at a time, and loosen the limits once the first threads come back the way you want.
Set alongside two related moves, the layering becomes clear: the Claude Code cloud sessions covered earlier decide where a task runs, and Managed Agents dynamic workflows decide how one job is split across many agents. Projects add the layer on top: when work keeps arriving over weeks, who receives it, who watches it, and who reports back to you.
