50 First Dates with AI
Current AI memory architecture is anterograde amnesia — and we're all Henry Roth, recording VHS tapes every night so our agent can wake up functional tomorrow.
The Talk
Every morning, your AI wakes up and doesn't remember yesterday.
We built agentic systems with anterograde amnesia — they can't form new long-term memories. Instead, we record VHS tapes (context windows) every night and hope the agent can reconstruct enough context to be functional.
This isn't a technical limitation. It's an architectural choice we keep making because it's easier than building systems that actually learn.
In this keynote, I show why the "50 First Dates" problem is the single biggest blocker to truly agentic AI — and how to build systems that compound instead of reset.
What You'll Learn
- Why your AI's 'memory' is actually just replay, not learning
- The three-layer memory architecture that actually compounds
- How to move from VHS tapes to persistent, evolving agents