Claude Code projects can burn your limits faster

Two threads edit the same lines, and Claude does not do anything clever about it: the overlap comes back as a merge conflict, like any other pull request. That is part of the design of the rebuilt Projects in Claude Code, announced on X and in beta from today for select Pro and Max subscribers on cloud sessions.

At a glance

• You brief Claude once in a project chat, and it scopes the request, delegates it to parallel cloud threads, reviews the outputs and assembles the result while your laptop is shut.
• Each thread is a full Claude Code cloud session running on its own branch and its own copy of the repo, and each can split its share further with subagents, loops and workflows.
• Because every thread is a full session, projects reach usage limits faster, and no thread can touch files, tools or an internal network on your own machine yet.

If you have not been following the plumbing underneath this: according to Anthropic's release notes on claude.com, Claude Cowork arrived on web and mobile alongside desktop on 7 July 2026, starting with the Max plan. The same notes describe shared memory across chat and Cowork, listed under Topics in Settings > Memory and editable there, on by default for Free, Pro and Max and off by default for Team and Enterprise. Projects now sit on top of both.

Each thread is a Claude Code session on its own branch

You start a project by picking a goal and the repo or context, and Claude opens with a list of work it can take on right away. From there you set the cloud environment, connectors, plugins, instructions and model. The coordinator scopes each request, hands it to threads, coordinates them in parallel, reviews what comes back and assembles the finished result.

Anthropic's examples are ordinary maintenance work. Set a goal to cut your checkout p75 latency, then ask Claude to profile each endpoint, test optimizations and open PRs in parallel threads. Or connect your API, web and mobile repos to retire a deprecated v1 endpoint: a thread per repo migrates the callers, runs the tests, opens PRs, and Claude tells you which ones need to merge first.

Under the hood each thread is a cloud session with its own branch and its own copy of the repo, a bit like handing three people a photocopy of the same book. The coordinator keeps the work organized, and where two threads touch the same code the overlap surfaces as a merge conflict. Threads can also split their assignment further using subagents, loops and workflows.

Every thread writes into one shared project memory

Memory is the part that changes how a project feels over a week. Every thread adds to a shared memory and draws from it, so Claude can hold that the release moved to Friday, why the export was dropped, or who to check with before anyone touches the billing service. Anthropic says this cuts the need for elaborate prompt engineering.

It also picks up your working and communication style, and you can tell it how often to check in, how readily to start new threads and how detailed each update should be. Alongside memory, a project library collects the files you add and the artifacts Claude produces, so new work can build on past efforts.

The Overview panel lists what is waiting on you. You can steer from the main project chat or drop into a single thread to examine the details, from a phone if that is where you are.

The beta excludes anyone who already has a project

Access opened today for a narrow slice: Claude Pro and Max subscribers who use cloud sessions in Claude Code and have no existing projects on web or desktop. If you are on Pro or Max and do not have it yet, there is a waitlist at claude.com/form/projects.

Existing projects on Pro and Max keep working exactly as they do today. Anthropic says it will upgrade them as the rollout expands to chat and Cowork, so nothing you have already built has to move now.

Threads run in the cloud only. They keep working while your computer is offline, and for the same reason they cannot reach files or tools on your machine, or anything on your internal network.

Several full sessions at once burn limits faster

Because a project can run several threads at once and each one is a full Claude Code session, projects reach usage limits faster. There is a project-specific usage view, and you can pick the model and effort level separately for the coordinator chat and for the worker threads.

Augmentcode, writing about agent orchestration, cites Anthropic's figure of roughly 15 times as many tokens for multi-agent systems as for chat interactions, the usual explanation for why orchestration costs more. Augmentcode also cites failure-taxonomy research finding that coordination failures account for 36.94% of all failures across AutoGen, CrewAI and LangGraph.

No numbers are attached to the limits claim itself, and the only timing given for local threads is «very soon». In our view the honest part of the design is that merge conflict: the coordinator organizes the work but does not pretend to arbitrate two threads editing the same lines, and leaves the resolution where developers already deal with it.

Where the rollout goes next

Anthropic says access widens to more Claude Code users on Pro and Max over the coming week, with updated projects across the rest of Claude, plus Team and Enterprise plans, after that. Old projects get upgraded as the rollout reaches chat and Cowork. The dated part stops there: threads running on your own machine, next to your local tools and behind your network, are promised very soon and nothing more precise.