Claude Code Projects Now Splits Itself Into Parallel Cloud Threads That Keep Running After You Close Your Laptop

Anthropic announced a full redesign of Projects in Claude Code on September 17, 2026 — the feature that used to be little more than a shared folder of files attached to one conversation. The new version turns a single ongoing chat into a coordinator that can spin up multiple working sessions at once, each one capable of writing code, opening pull requests, and running tests without you watching it happen.

What Actually Changed in Projects?

Anthropic’s own framing is blunt: the old Projects was a folder — some files plus one chat. The new one is a conversation. Instead of manually dividing work across sessions and stitching the results back together yourself, you describe what needs to get done, and Claude scopes the request, delegates the work, coordinates parallel threads, reviews the outputs, and assembles the finished result.

You can steer progress throughout — including from your phone — and the work keeps going after you step away from your computer. Anthropic gives two examples of what this looks like in practice: setting a goal to reduce an app’s checkout p75 latency, then having Claude profile each endpoint, test optimizations, and open pull requests in parallel threads; or connecting API, web, and mobile repositories and asking Claude to retire a deprecated v1 endpoint, with one thread per repository reporting back which pull requests need to merge first.

How Do the Coordinator and Threads Actually Work?

The architecture has two layers. The project conversation is the coordinator: it reads what you send, answers quick questions directly, and starts threads when there’s real work to delegate. It sees what threads report back, not every step they take along the way.

Threads are the workers. Each one is a full Claude Code cloud session running on its own git branch and its own copy of the repository, so multiple pieces of a project can move forward at once without one thread overwriting another’s changes. A thread opens a pull request when the work calls for one and can watch that PR with auto-fix enabled, pushing fixes when CI fails. If two threads happen to touch the same code, the overlap surfaces as an ordinary git merge conflict — not a special failure mode, just the same conflict you’d get from two human contributors.

Threads can delegate further, splitting their own assignment into pieces using subagents, loops, and other workflows when a task is large enough to warrant it.

What Do Projects Remember?

Every new thread inherits standing context automatically: the project’s connected repositories and uploaded files, project instructions of up to 16,000 characters, and shared project memory that builds over time. Anthropic’s example of what gets remembered is mundane and specific — things like “the release moved to Friday” or “who to check with before touching billing.” This is a separate mechanism from the shared memory Anthropic rolled out across Chat and Cowork earlier this year, but it follows the same underlying idea: context that used to evaporate at the end of a session now persists and compounds.

Projects also include a library that collects both the files you upload and the artifacts Claude produces along the way, so new work can build on what earlier threads already did instead of starting from nothing.

Who Has Access, and What Does It Cost to Run?

Beta access started September 17 for a narrow slice of users: select Claude Pro and Max subscribers who already use Claude Code cloud sessions and don’t have existing projects on web or desktop. Anthropic says access will widen to more Pro and Max users on Claude Code over the following week, with updated projects reaching the rest of Claude, plus Team and Enterprise plans, after that. Existing projects on Pro and Max keep working as they do today until they’re upgraded.

The cost trade-off is direct: because a project can run several threads at once, and each one is a full Claude Code session, a project can burn through plan usage limits meaningfully faster than a single chat. Anthropic exposes a per-project usage tab and lets you set the model and effort level separately for the coordinator conversation and for the worker threads — so you might run the coordinator on a cheaper, low-effort setting while giving threads doing the actual work a more capable model. Reporting on the beta also notes an enforced cap of 200 new threads per day across a user’s projects, with a thread that hits a usage limit simply waiting and resuming on its own rather than failing outright.

One real limitation for now: threads run in the cloud only. Anthropic says running threads locally — alongside your own machine, tools, and network — is “coming very soon,” but isn’t available at launch.

Why Does This Matter Beyond Claude Code?

This is the second major Claude Code change in as many months, following Claude Fable 5.1’s launch with its own coding and cost improvements. Taken together, they point at the same bet: Anthropic thinks developers want to hand off entire goals, not individual prompts, and are willing to trade tighter manual control for a system that keeps working unattended. That is a meaningfully different product shape than a chat window, and it puts Claude Code more directly in competition with OpenAI’s own coding-tuned models on the same axis: how much of a real engineering task can the model own end to end, not just how good a single response is.

The trade-offs are also worth taking seriously before treating this as a pure win. A tool that keeps running after you close your laptop, opens its own pull requests, and burns plan usage in the background is a bigger blast radius for a mistake than a single chat session — which is presumably part of why access is still limited to a narrow beta group rather than rolled out broadly on day one.

Frequently Asked Questions

Is the new Claude Code Projects available to everyone?

Not yet. Beta access started September 17, 2026, for select Claude Pro and Max subscribers who use Claude Code cloud sessions and don’t already have existing projects on web or desktop. Anthropic says access will widen over the following week, with Team and Enterprise plans getting it later.

Can Claude Code Projects run on my own computer instead of the cloud?

Not at launch. Every thread currently runs as a cloud session. Anthropic says local execution, running threads on your own machine alongside your existing tools and network, is coming soon but was not available when Projects launched.

Does running parallel threads cost more than a normal Claude Code session?

Yes. Each thread is a full Claude Code session, so a project running several threads at once consumes plan usage faster than a single chat. Anthropic provides per-project usage tracking and lets you set different models and effort levels for the coordinator versus the worker threads to help manage cost.

What happens if two threads edit the same code?

It surfaces as a standard git merge conflict, the same as if two human developers had edited the same file on different branches. It is not treated as a special error case.

How is this different from just running multiple Claude Code sessions manually?

Previously, splitting work across sessions meant manually dividing tasks, tracking each session yourself, and combining the results at the end. In the new Projects, you describe a goal once, and Claude decides what becomes a thread, delegates it, monitors progress, and assembles the outcome, while also building shared memory that every new thread can draw on.

Sources: Anthropic, MarkTechPost.

About the author

Kai Williams

Kai Williams has been in marketing for years, with a long background in SEO before AEO had a name. He stepped into Answer Engine Optimization the moment AI started reshaping how people search, and has been tracking the shift ever since. At Prompt Insider, he covers AEO, AI marketing, and the future of search, breaking down what is changing and what brands need to do about it.