Claude Code now
keeps working after you close the lid.
Anthropic has redesigned "Claude Code Projects" and reopened it in beta. A single conversation now fans out into multiple cloud-resident agent threads running in parallel — and they keep going after you close your laptop. Here's what actually changed, and for whom it matters.
No longer built around
"close it, it stops"
Anthropic announced that on September 17, 2026 it redesigned the "Projects" feature inside Claude Code and reopened it in beta (MarkTechPost, The Verge).
Until last month, Claude Code was fundamentally a tool you ran one local session at a time on your own machine. Close the terminal or the laptop, and the work stopped — anything long-running meant leaving the machine on.
The new Projects turns a single, ongoing conversation into a "coordinator" that receives a goal and breaks it down into multiple worker threads, run in parallel in the cloud — a two-tier "coordinator + worker thread" architecture (The Decoder). Each worker thread has its own branch and its own copy of the repository, and is itself a full Claude Code cloud session. The important shift here isn't that it's "faster" — it's that the unit of agent execution has been decoupled from the human's own session.
The Decoder frames this redesign not as a one-off feature but as part of Anthropic's ongoing push to move Claude Code toward more autonomous coding. As more of these parallel-agent workflows ship, coding assistance shifts its center of gravity from "a tool you operate hands-on" toward "hand off a goal, come back for the result."
| Claude Code, before | Claude Code Projects (new) |
|---|---|
| One session at a time, on your machine | Multiple threads, running in parallel in the cloud |
| Stops when the laptop closes | Keeps going after you close it |
| Progress visible only on your local screen | Check and steer progress from chat or your phone |
| A single working branch | A dedicated branch and repo copy per thread |
Describe what needs doing, and Claude directs the parallel threads.
They keep working after you close your laptop.
The coordinator hands out the work
A single conversation acts as project manager, routing tasks to multiple cloud sessions.
State a goal in one conversation
You describe a broad goal — e.g. "fix the API rate limiting, migrate the web app, and check the mobile side" — in a single, ongoing conversation.
The coordinator scopes and assigns it
Claude scopes the request and routes tasks to new or existing worker threads. Each thread gets its own branch and its own copy of the repository.
Threads build, test, and open PRs in parallel
Each thread makes code changes, runs tests, and opens a pull request in the cloud. You can track progress and steer it from chat or your phone.
Who benefits, and how
The upside — and the fine print — differ by role.
Engineers
Push several independent tasks forward at once — an API fix, a web migration, a mobile check — and stay focused on design and the tricky reviews yourself.
Business & leadership
Team throughput can rise, but every thread is billed as a full session's worth of usage — so cost-effectiveness is still something to judge from real team usage, not assume.
PMs & program leads
See several tasks moving in parallel from one chat and your phone, without pinging someone at their local screen for a status check — handy for a quick pre-standup read.
What to do next
The beta began rolling out on September 17, 2026 to a select group of Claude Pro/Max subscribers already using cloud sessions, with access widening over the coming weeks and Team/Enterprise access to follow. Support for running it locally is reportedly in the works too.
Three concrete moves for eligible users. First, try it on one real task — pick independent pieces of work, like an API fix and a docs update, so you can actually measure the effect. Second, check your usage budget up front: worker threads default to Opus at high effort, separate from the coordinator conversation's own model and compute tier, so don't assume you can run every thread at full power indefinitely. Third, route work that's unlikely to collide — if multiple threads touch the same files, you're back to ordinary merge-conflict resolution by hand.
This isn't unqualified good news
The first limit is cost. Because every worker thread counts as a full session, usage burns faster the more threads you run in parallel — more threads isn't simply better.
Second, availability is still narrow. This beta targets Pro/Max subscribers already using cloud sessions; those with an existing Projects setup on web or desktop are reportedly excluded for now, and local execution isn't available yet either.
Third, parallelism doesn't suit every task. Split a tightly coupled change across threads and you'll still end up doing the merge coordination yourself. The clearest wins are for work that's naturally separable — profiling different endpoints, or migrating API, web, and mobile in parallel. Before handing off a task, it's worth a moment to ask whether it will actually collide with another thread.