Claude Just Shipped Native Memory. Here's What I Actually Think (Building a Memory Startup)

Anthropic shipped native Memory for Claude this week. If you're building in this space like I am, that's the kind of news that makes your stomach drop for about ten minutes.

Then I actually used it, and the reaction changed.

It's genuinely good for what it does: Claude now remembers project details and preferences without you re-explaining them, and you can even import history from ChatGPT. For someone who only uses Claude, this closes a real gap.

But it's still fundamentally single-vendor. The import feature is a one-time transfer, not a live sync, so the moment you're back in ChatGPT or Gemini for a different task, you're starting fresh again there. It solves forgetting inside Claude. It doesn't solve fragmentation across tools, and almost nobody I know uses just one AI anymore.

That's the part I'm still building for. Not betting against native memory, I think every platform will have some version of it within a year. Betting that "context that follows you across tools" stays a real, separate problem even after every vendor solves their own corner of it.

Curious if other founders building on top of foundation-model features have gone through the same "oh no, then oh okay" cycle this week.