Anthropic fixed the most frustrating thing about using Claude across chat and the Cowork app by finally making memory shared between both, so you do not have to keep explaining your projects and preferences every time you switch. This is one of the most frustrating aspects of any public AI with multiple service types (e.g., chat and coding).
I have a process for writing “resume session” documents (.md files): I end one chat or coding session by having my AI create this document, then start a new session referencing the resume session document. This was my hard-coded memory, which seemed to work pretty well. And since I could read these session resume documents myself, it refreshed my own memory.
The game of leapfrog continues between the major AI players. I would guess that OpenAI is working on something similar so they aren’t left behind. I will be testing this out within my Claude environment. But treat AI like a child. If you do, that means you will have to continue to repeat and clarify yourself, but hopefully a bit less frequently. Fingers crossed.