A team lead pastes a bug list into 1 conversation on Monday morning, and by lunch 4 threads are working on it at once. That is the new shape of Claude Projects. You describe the work once, Claude starts a thread for each task, and the threads run in parallel in the cloud while your laptop sits closed. The change is real, and it comes with a meter attached. The project documentation states the meter in plain words: a project draws on the same plan limits as your other sessions, and running threads at once uses those limits faster.
The direct answer: Claude Projects is now 1 coordinating conversation that splits work into parallel threads, each with its own context, its own branch and its own report back. It is in public beta on Pro, Max and Ultra plans, starting with a narrow set of Claude Code users, and Team and Enterprise plans follow later. A Canadian team that gets access should set 2 rules before the first batch: a cap on threads running at once, and a propose-first rule until the first threads come back clean.
- What changed in Projects
- The 2 rules to set before the first batch
- Who can use it today, and who waits
What changed in Projects
A project used to be a folder of files and chats. It is now a stream of work with a coordinator. You send tasks to 1 conversation, in batches or as they arrive. Claude routes each task to a new thread or to the thread already working in that area. Each cloud thread starts with the project’s instructions and memory, works on its own branch, and opens a pull request when the work calls for one. The Overview pane shows which threads finished and which thread is waiting on you.
The detail that changes office habits is the default setting. A new project runs every thread on Opus at high effort unless someone changes it. That default draws on the plan fastest, at the exact moment a team is most likely to send a large batch to see what happens. The bill arrives as a plan limit reached mid-week, with the routine work still queued behind the experiment.
The 2 rules to set before the first batch
Rule 1 is a thread budget. Decide how many threads may run at once, and write the number into the project instructions where every thread can read it. Start with a few at a time. The maker’s own guidance says the same: ask Claude to propose threads before starting them, and drop the limit only after a few threads come back the way you want. Give your AI agent an 8-line job description before access set the wider rule on this site: access and limits get written down before the agent starts, in lines a person can check.
Rule 2 is a review rule. A thread that finishes is a claim, and a claim needs a check against the record it should have left. Pick one project marked ‘done’ and ask for the record it should have left behind gives you that check in audit form. Project memory needs the same discipline as thread count. Your agent’s memory needs an editor, not a bigger limit covers the memory side, and Notion agent Memories are private to each user. Decide what they may keep covers what a shared memory should hold at all. A project that remembers a wrong call will hand that call to every thread that follows.
Who can use it today, and who waits
Access is narrow at launch. The beta reaches select Pro, Max and Ultra subscribers who use Claude Code cloud sessions, and expands to more users on those plans as the rollout moves. Team and Enterprise plans come after that, so a Canadian office on a team plan cannot switch the feature on today. That gap is useful: a team lead gets time to write the thread budget and the review rule before access lands.
The recap is short. Projects now coordinates parallel threads from 1 conversation, each thread spends from the same plan limit, and the default setting spends fastest. The budget and the review rule cost nothing to write today.
Do this week, before access reaches your team: open a blank project brief, write your thread cap and your propose-first rule into it, and name the person who reviews finished threads. No hands-on test was run for this piece, so your first capped batch is the test, and its limit report is the number to bring back.