Anthropic launches Claude Code Projects, an ‘always On’ conversation that remembers and delegates your long Running dev work

Pippa Okay, that’s the part that gets me. If you’ve ever had to restart a coding session just to remind the agent what it was doing, Claude Code Projects is basically Anthropic admitting that the real job is continuity, not a fancier prompt box.

Tyler Yeah, and that’s the right admission. The annoying part of long-running dev work has never been writing one good prompt. It’s that the work fragments, the context drifts, and then you become the scheduler, the note-taker, and the poor soul reconciling branches at the end.

Pippa Exactly. And this isn’t just “remember my last chat.” Anthropic is saying the project sits above the threads, keeps shared memory, and can decide whether a new request goes into an existing thread or spins up another one. That’s a much more product-shaped answer than just saying, well, the model has a bigger window now.

Tyler Right, and the architecture matters. Each worker thread is its own cloud session on its own branch, so the concurrency is real. But the clever bit is the coordinator layer: threads do the work, Claude directs it. That’s the thing that makes it feel less like a pile of parallel chats and more like an actual operating layer for a project.

Pippa And that’s the adoption story, too. I can see this landing for teams maintaining a storefront, a CMS, a procurement system, anything that never really ends. You don’t want to re-brief an agent every time a ticket shows up. You want it to remember that the release date moved, that a feature got dropped, and that one service needs sign-off before anyone touches it.

Tyler Mm-hm.

Pippa The nice part is that Anthropic’s not pretending parallelism alone is the magic trick. They’ve got an overview panel for attention items, they say you can steer threads from a phone, and the project can carry its own environment, connectors, plugins, instructions, and model settings. That’s the stuff that turns it from demo candy into something a team might actually keep open.

Tyler Yeah, though the name is doing them no favors. They already have Claude Projects, which is the general workspace for chats and files and knowledge. Now they’ve got Claude Code Projects, which is really about coordinating agentic software work across sessions. That’s a pretty easy way to confuse people who are not living inside Anthropic’s naming spreadsheet.

Pippa Anthropic really said, what if we made the second thing sound exactly like the first thing. Very on brand for a company that keeps making infrastructure you have to squint at before it clicks.

Tyler No, seriously, that’s the one product risk I’d flag. The idea is clean, the naming is muddy. If I’m a developer skimming this, I have to do a little mental unpacking before I even get to the good part. And the good part is real: shared project memory across threads is more useful than just spawning a bunch of isolated agents and hoping human memory fills the gaps.

Pippa That’s why I think the chief-of-staff framing is actually decent, annoying as it sounds. You hand it a messy objective, it routes the pieces, and you keep talking to one coordinator instead of babysitting every subtask. That’s a better user story than “congratulations, you now have six agents to manage.”

Tyler Yeah, and the comparison set is the point. OpenAI’s Codex workspace already normalized parallel agents. GitHub is coming at it from the repo and lifecycle side. Anthropic’s angle is a persistent conversational coordinator with memory above the worker threads. That’s not revolutionary, but it is a sharper interface choice if the thing you care about is not losing the plot halfway through.

Pippa Which is basically the whole product bet. The question isn’t whether an agent can fix a bug. The question is whether it can stay useful when the work drags on, the requirements change, and the human comes back three sessions later asking, wait, what did we decide about this endpoint?

Tyler Right. And I’ll be a little skeptical in the boring way here: if they nail the coordination, great. If they don’t, you just created a more elaborate place for context to rot. But I do think this is the right direction. The field keeps moving from coding agents to agent managers, whether everybody likes that phrase or not.

Pippa That’s such an Exploring Next take, by the way. Deeply cursed, probably true.

Pippa I’m genuinely curious whether this becomes the default way people think about Claude Code, because the workflow story is strong. Not “use this for one prompt,” but “keep the project alive and let it carry the work.” That’s the kind of thing teams notice when it saves them from re-litigating the same setup over and over.

Tyler And it’s the kind of thing you can only really judge once people try to live with it. The launch is still in beta, it’s going out first to select Pro and Max users on cloud sessions, then broader. That rollout path makes sense. If the coordination layer is good, people will feel the difference fast. If it’s clunky, they’ll know almost immediately.

Pippa Yeah. Which is why this one feels less like a flashy model drop and more like Anthropic trying to own the layer above the model. Okay, that’s enough intellectual honesty for one Wednesday.

Tyler We are absolutely doing this on a Wednesday, yeah.

Pippa Go see if Claude can keep your project straight better than you can. If it can, that’s honestly embarrassing for all of us, but fine.