The part that catches my attention isn’t that Anthropic is adding memory. Everyone is doing some version of that now. It’s that Claude Chat and Claude Code are apparently being tied together so the model carries one shared memory across both. That’s the bit I’d actually want to test, because if it works well it removes a lot of friction, and if it works badly it becomes a very efficient way to import one context’s nonsense into another.
I can see why this is attractive for people using Claude as a real work tool. A coding assistant that remembers your preferences from chat, or a general chat assistant that remembers the conventions of your codebase work, sounds genuinely useful. The ideal version is not “the model remembers everything,” which would be creepy and brittle. It’s more like a persistent working style: names, project assumptions, the tone you like, the kinds of answers you ignore. That’s the kind of memory that saves time without demanding that you restate your life every session.
But memory is also where assistant products start pretending they know you better than they do. The source mentions controls around sensitive topics like health and beliefs, which is the right instinct, but it also hints at the deeper problem: once you give a model a memory layer, you’re no longer just asking it to answer current prompts. You’re asking it to carry forward inferences. That can be helpful, but it can also calcify mistakes. If Claude decides you prefer one pattern, or misunderstands something important about your work, that wrong guess can haunt every later interaction until you notice it.
I’m also curious how clean the boundary is between “chat” and “Cowork.” If the memory is truly shared, then users need to understand exactly what gets remembered, where it gets applied, and how easy it is to reset. In practice, these systems often sound more controllable than they feel. The docs may say one thing, but the real question is whether you can predict the behavior well enough to trust it in daily use.
So yes, this feels like a sensible product move. It’s also the kind of change that only looks boring until it silently shapes how much people rely on Claude. That’s the real test here: not whether memory exists, but whether Anthropic can make it feel like a helpful notebook instead of a surveillance diary.
Reference: Anthropic gives chat and Cowork one memory - The New Stack
