Claude Code Projects turns a codebase into one conversation
One agent per project handles memory and spins off subagents, and the work keeps running after you close the laptop.
Projects is rolling out in Claude Code on desktop and web, and the pitch is that you stop opening sessions. ClaudeDevs describes a project as one conversation with Claude that splits the work into threads itself, runs them as parallel cloud sessions, passes context between them, and keeps going when you leave. It is in beta for select users.
Boris Cherny, posting as @bcherny, says the change is not cosmetic. He writes that Projects have changed not only how he interacts with Claude but how he codes, that he now just sends thoughts as they come, Claude splits them into threads, and the project remembers how he works. He says it is where he does a ton of his coding now, and in a separate post that Projects are how he writes a lot of his code these days.
I stopped managing sessions.
What is actually running
Thariq (@trq212) gives the architecture. He says Projects brings the architecture of Claude Tag to Claude Code, with one agent per project that manages memory and spins off subagents for tasks. He adds that you can ask it to be proactive or to do things on a schedule, and that it feels a lot nicer than a bunch of sessions.
For working engineers, that one memory agent is the whole bet. Every session you open by hand is a decision about what context the model sees, which files matter, which prior conversation is relevant, which thread to resume. Projects takes that decision away and makes it the agent's problem. If the memory agent picks well, the overhead of curating context disappears and the parallel threads are free. If it picks badly, you get a confident agent working from the wrong slice of your codebase, and the thing that used to be a bad prompt is now a bad prompt you did not write.
The evidence so far is two people at the company that shipped it saying they like using it, which is worth something and is not the same as a result on someone else's codebase. Cherny's post includes a screenshot he labels as his actual prompts for the Claude Code CLI the previous day. Nobody outside has published a comparison against manually managed sessions yet.
It has one agent per project that manages memory and spins off subagents for tasks.
Have the good opinion, not the loud one
One email a day. Under three minutes. What working engineers found out about coding with AI.