If your firm has used Claude Projects, you probably set them up like matter folders. You dropped in the engagement letter, the complaint, a few key memos, and a page of instructions about tone. Then you opened chats inside the project. Each chat could read the folder, but each one mostly started from scratch. If you spent an hour in one chat working out the theory of a motion, you carried that thinking into the next chat yourself.
Anthropic has rebuilt Projects around threads. It announced the change on September 17 and is rolling it out in stages. One of my own projects switched over in early October. It greeted me with “Hi Steve, welcome back. Tell me what’s next.” My old chats were listed as threads, and what Claude remembered from the old project came over with them.
Then I gave it two unrelated jobs, a few minutes apart. Each one got its own thread, and Claude worked on both at the same time. Later that day I typed a follow-up on the second job into the main project chat. I never said which thread it was for, and Claude passed it to the right one anyway.
A project used to be a place where you kept things. Now it’s a conversation where you hand out work. For a law firm, that changes how you set projects up and how you supervise what comes out of them.
How the new Projects work
You tell the project what you need in plain English, and Claude acts as a coordinator. Quick questions get answered right there. Bigger jobs go to threads. Each thread is a full working session in the cloud that reads your files and reports back. It keeps going after you close your laptop.
Underneath it all is one shared memory. Every thread writes to it and reads from it, so a fact you mention in one thread can shape the work in another. There’s also a Library that collects the files you add and the documents Claude produces, so finished work stops getting scattered across a dozen chats.
I think this fits legal work better than the old setup did. A matter is a pile of related tasks that lean on the same facts. When the chats couldn’t see each other, you were the one holding the file together in your head.
What changes for a litigator
Here’s a hypothetical. You’re defending a breach of contract case, and this week you need a chronology built from the document production and a first-draft outline for the CFO’s deposition.
In the old setup, you’d open one chat for the chronology and another for the outline, and you’d paste the same background into both. Then on Wednesday you learn the CFO left the company in March. Now you have to remember which chats need that fact and tell each one.
In the new setup, you ask for both in one message and the coordinator starts two threads. When you mention the CFO’s departure, it goes into project memory and both threads pick it up. The same goes for a trial date that moves or a client who takes an argument off the table. You say it once.
You’ll also brief it differently. Until now, the skill was writing a careful prompt for each chat, and a lot of lawyers got good at it. Anthropic’s announcement compares the new approach to briefing a chief of staff. For lawyers, the closer match is a partner handing a matter to a senior associate. You give the goal, the client’s priorities, the deadlines, and the things you don’t want done, and then you let it organize the work.
Who has it now
Anthropic started with select Pro and Max subscribers using Claude Code. Its help center says existing projects on Pro and Max plans get upgraded as the rollout reaches the regular Claude app, and mine already has been. Team and Enterprise plans come after that. Anthropic hasn’t given a date.
So if your firm runs Claude on a Team or Enterprise plan, your projects probably haven’t changed yet. But a lawyer at your firm doing client work on a personal Pro or Max account may already be in the new version. If that’s happening, you have a bigger policy problem than this upgrade.
Sharing is the open question for firms. My upgraded project is private to me, and I can’t add anyone to it. Classic projects on Team and Enterprise plans can be shared with colleagues, and Anthropic hasn’t said how that will work once the new version reaches those plans. I wouldn’t rebuild a practice group’s shared projects around it until it does.
The risks I’d watch
Shared memory worries me most. Under the old setup, it was easy to get away with a loose project, say one called “Employment matters” with files from several clients in it. That was never a good idea, and it’s a worse one now. Every thread writes to the same memory, so a fact from one client’s matter can show up in work for another. The fix is simple and a little tedious. One project per matter, or at the very least one per client.
Then there’s supervision. Threads keep working after you walk away, and the coordinator decides how to split up your request. ABA Formal Opinion 512, issued in July 2024, says managing lawyers need clear policies on generative AI. Supervising lawyers have to make reasonable efforts to see that people follow them. A tool that works overnight doesn’t change those duties. But it makes it easier for a finished draft to sit in the Library before any lawyer has read it.
Cost is the last one. Anthropic says each running thread is a full session, so projects burn through plan limits faster than single chats. You can see usage for each project, and you can set the model and reasoning level separately for the coordinator and the threads. I’ve already turned the reasoning level up on one thread when I wanted more careful work. That control helps, but someone at the firm still has to watch the meter.
The fair objection is that most lawyers just want help with a letter, and all of this sounds like overkill. I understand that. The coordinator still answers simple requests on its own without starting a thread, so you can ask for one letter and get one letter. What goes away is the habit of opening a fresh chat for every task and being the only one who knows how they connect.
What to do Monday
Audit your current projects and split any that mix clients or matters. Do it before the upgrade reaches your accounts, because once threads start writing to shared memory, untangling two clients’ facts gets harder.
Rewrite each project’s instructions as a partner’s brief. Name the client and the goal, list the deadlines and the arguments to avoid, and say who reviews the work.
Pick one person per matter to check the Library and the project memory every week. That person corrects anything wrong in memory and signs off on anything that leaves the firm.
Treat the new project like a new associate on the matter, and supervise it the same way.
Thanks for reading! This is a big change for all of my clients. I’ve been telling everyone I train go use projects so you can ensure that memory and information doesn’t bleed from one client to another. I think this is still the right advice but until we see the final Team and Enterprise approach - I’m going to reserve judgement.
If you enjoyed this, please share it with others!