What jumps out to me is not the “memory” bit itself. It’s that Anthropic is finally making Claude act a little more like one continuous system instead of a bunch of half-connected surfaces. If you’ve ever bounced between a chatbot, a browser extension, and an agent workflow and had to re-explain yourself each time, you know how annoying the current state of these tools is. That friction is not a small UX papercut; it’s the thing that makes “agentic” products feel fake.
The cross-over between chats and Cowork is the part I’d actually test first. If Claude really carries context from my normal conversations into a task-running agent, that’s useful in a very practical way. It means less repetition, fewer prompts that start with “as I told you before,” and fewer little failures where the agent behaves like it woke up with amnesia. Still, I’d want to see how disciplined the memory system is in practice. A lot of these features sound smart in demos and then become a junk drawer of stale assumptions, half-remembered preferences, and weird one-off facts that keep resurfacing at the wrong time. That might be the real risk here.
The sensitive-topics toggle is more interesting than it first looks. Anthropic is basically admitting that “sensitive” is not a universal category, and that’s true. For some people, remembering politics or health context is a privacy nightmare. For others, forgetting it makes the assistant much less useful. I think that framing is sane. What matters is whether the controls are actually legible enough for users to trust them, and whether Claude is conservative enough by default. The opt-in approach sounds right to me, but the proof will be in the details: what gets surfaced in the memory UI, how easy it is to delete, and whether it ever quietly infers more than it should.
I do think the default-enabled memory is where Anthropic is making the biggest bet. Memory is one of those features that becomes valuable only after it has accumulated enough history to feel personal, but that same accumulation is exactly what makes people uneasy. There’s a tension here that every assistant vendor is going to have to live with: the more useful the system gets, the more it has to know; the more it knows, the more it needs to earn trust. Anthropic seems to be leaning into that tradeoff instead of pretending it doesn’t exist.
Reference: Claude's memory now works across both chats and Cowork sessions
