Claude Code Can Now Run Work in Parallel — And Spend Your Plan in Parallel Too

TL;DR: Claude Code has a new Projects feature in beta: you describe a body of work, and Claude fans it out into parallel cloud sessions that keep running after you shut your laptop. It’s genuinely useful. It also spends your allowance considerably faster, and every thread defaults to Opus at high effort — the most expensive setting available. Anthropic documents this plainly, including that there’s no cap on how many threads run at once, only 200 new ones a day. If you’re on Pro, change one setting before you try it.

What Claude Code Projects actually does

A project is one ongoing conversation where Claude acts as a coordinator. You paste in a bug report, a stack trace or a list of tasks, and it starts a thread for each piece of work — or hands it to a thread already working in that area.

Each thread is a full cloud session: Claude Code running on Anthropic’s machines rather than yours. Per The Register, each one gets its own git branch and its own copy of the repository, and the coordinator surfaces merge conflicts when two threads touch the same code. Anthropic’s summary: “Claude scopes the request, delegates the work, coordinates parallel threads, reviews the outputs, and assembles the finished result.”

The practical appeal of Claude Code Projects is real. Threads keep going after you close the laptop, you can check on them from your phone, and you stop doing the coordinating yourself — no repeating the same background at the start of every session, no checking which one stalled.

It’s in public beta on Pro and Max, rolling out first to accounts that have used cloud sessions and don’t already have projects in Claude chat or Cowork. Team and Enterprise don’t have it yet, and it’s web and desktop rather than the command line.

The setting to change first

Here is the fact that matters most, and it’s in Anthropic’s own Claude Code documentation rather than buried anywhere:

“A new project runs every thread on Opus at high effort, which draws on your plan fastest.”

The coordinator conversation runs at low effort. The threads — the things you’ll have several of at once — default to the most expensive combination on offer. You can change it: Project settings → General → Thread model and Thread effort.

Anthropic is equally direct about the consequence: “A project uses your limits faster than a single session does, and on a Pro plan in particular you should expect to reach your limit sooner on days you run one.”

That is a vendor telling you, in its own docs, that its new feature will burn through your plan. Worth crediting — and worth acting on.

Three things that spend money while you’re not looking

The documentation lists what draws on your plan, and three details deserve pulling out because none of them is obvious from the interface.

There’s no limit on how many threads run at once. In Anthropic’s words: “There’s no fixed number; Claude starts as many as the work calls for.” The only enforced ceiling is 200 new threads per day across your projects. And if you ask Claude in conversation to run only a few at a time, that’s a preference, not a cap — the docs say such instructions are “instructions Claude keeps to, not enforced settings, so a thread limit you give this way isn’t a hard cap.”

A thread that hits your limit doesn’t stop — it waits. “A thread that reaches your plan’s limit waits and continues on its own when the limit resets, so work you left running starts using your next usage window without a message from you.” Read that twice. Work you started on Tuesday can quietly begin spending Wednesday’s allowance before you’ve opened your laptop.

Idle threads wake up and spend again. A thread watching a pull request “wakes up and uses your plan again when CI fails or a review comment arrives.” You have to tell it to stop watching.

The reassuring counterweight, also from the docs: a project with no running threads, no watched pull requests and no new messages doesn’t use your plan while it sits idle, and neither does an archived one. And a project can’t spend past your plan limits at all unless you’ve separately turned on usage credits — “A thread can’t turn them on for you.”

Why this matters if you don’t code

Ten days ago we wrote about a class action alleging Claude Max’s “20x more usage” describes a five-hour window rather than the weekly cap. The central problem in that piece was that a multiplier tells you nothing if you can’t estimate what a unit of work costs.

Claude Code Projects makes that problem sharper, because it changes the unit. Until now, roughly, one session was one thing happening. Now one instruction can become eight things happening, each a full session, each on the priciest model setting by default.

For a developer that’s a trade they can feel. For someone building an app as a means to an end, three specific risks:

• You won’t see it coming. The feature’s whole appeal is not having to watch it. That is also precisely why the bill arrives without warning.
• The default is the expensive one. Most people never change a default. Here the default is Opus at high effort on every thread, and Anthropic says so outright.
• In Claude Code, parallel doesn’t mean faster for you. If your project is one small app rather than a migration across many files, you may get the consumption of eight sessions for the usefulness of one.

There’s also a pattern worth noticing. Cursor shipped its own Projects feature with a coordinator and subagents on 10 September, a week before this. Two of the tools in our comparison have, within days of each other, moved from “an assistant you watch” to “a manager running work you don’t.” That’s the direction, and the pricing implications arrive with it.

How to try Claude Code Projects without regretting it

1. Change the thread model before your first project. Project settings → General. Opus at high effort is the default; something cheaper is almost certainly enough for a first run.
2. Ask Claude to propose threads before starting them. The docs suggest exactly this, and it’s the difference between reviewing a plan and discovering one.
3. Start with one real task, not your whole backlog. Watch what a single thread costs before handing over ten.
4. Check your usage percentage before and after. The same measurement we recommended for the Max plan works here, and it’s the only honest number you’ll get.
5. Tell threads to stop watching pull requests when you’re done with them, or they’ll wake up and spend again.
6. Pause or archive a project you’re not using. Idle projects don’t draw on your plan, but only once nothing is running or watching.

Who should care (and who shouldn’t)

• On Pro and building something substantial: the most exposed group, and the one Anthropic explicitly warns. Step 1 is not optional.
• On Max 20x running Claude Code hard: this is arguably what you’re paying for. Still worth measuring a project against a normal session once.
• Building a small app or a landing page: you probably don’t want this yet. The coordination overhead is the product, and you may not have enough parallel work to justify it.
• Learning: skip it. Watching one session work is how you learn what the tool does; eight at once teaches you nothing and costs more.
• On Team or Enterprise: not available to you yet.

Our take

We think this is a good feature and a genuinely risky default, and both halves of that sentence matter.

Anthropic’s documentation here is better than the industry norm — it says plainly that threads draw on your plan fastest, that Pro users should expect to hit limits sooner, and that a thread limit you ask for isn’t really a limit. Very few companies write the sentence “this will use up what you bought, faster” about their own new feature. After a fortnight in which we’ve covered a lawsuit over usage claims and an AI that concealed problems from its user, a vendor being this direct deserves saying so.

But defaults are decisions made on the user’s behalf, and this one sends the most expensive model at the highest effort into an unbounded number of parallel sessions, for a feature whose selling point is that you don’t watch it. Anyone who doesn’t read the settings page — which is most people — gets that configuration by accident.

The honest summary: Projects is a real improvement in how work gets organised, and a real change in how fast money leaves. Turn it on deliberately, change the thread model first, and run one task through it before you trust it with a day’s work.

Not sure whether Claude Code — or which plan — fits what you’re building? Take the 60-second Vibe Coding Tool Finder quiz →

FAQ