Last Thursday I gave two talks back to back, at AWS Summit Taipei and Agile Taipei. I didn’t expect both to be completely packed. Honestly, far more people came than I expected 😅 Thank you so much to everyone who made time to come, and it was really nice to see a lot of familiar faces in the audience.
This time I mainly talked about agent memory, which I’ve been researching and stumbling through lately.
People are increasingly aware of how important memory is for agents. But to put it bluntly, “memory” covers everything: the state of the current conversation, project knowledge, past task experience, how to carry things across sessions, even how to share them across a team. It all gets discussed in one big pile.
They all fly the memory banner, but compare Claude Code, Codex, OpenClaw or Mem0 and you’ll find each of them is really focused on something different.
What I find most interesting is that agent memory for an individual mainly solves continuity. But once that memory has to work at the team level, what you actually face becomes provenance, scope, review and conflict, and you even have to start dealing with how agents forget.
This was also my first time pulling together the research I’ve been digging through, along with the question Cairn really wants to answer, “how should people and agents share team memory,” into one complete talk. I got a lot of interesting questions, and it made me even more sure this topic is worth digging into 💪
I’ve read a pile of papers and product docs for this lately, and I’ll turn them into a few essays to share over time. Have you hit any memory-related pain points while building agents recently? I’d love to hear about them in the comments 🙌 And thanks again to AWS and Agile Taipei for having me~
