Giving AI Agents Memory That Learns — Jake Broekhuizen, LangChain
- Source
- youtube.com
- Author
- AI Engineer
- Date
Context
Jake Broekhuizen of the Labs team at LangChain argues that watching what an did does not make it better. Logs become memory only when a lesson is written into the agent reads on its next run. His AI Engineer World's Fair 2026 talk uses a financial services agent whose tone slid from advisory toward pushy as the example.
According to the talk description, he separates semantic memory (facts), episodic memory (past events) and procedural memory (instructions for how to behave), and says procedural memory moves behavior most. His loop is: read context, run, filter the evidence for signal, write it back. His lessons are that most traces stay history, caching needs care, and a human should review changes to key rules.
These are one practitioner's recommendations; the description reports no measured outcomes for the finance example.
- Agent — A system in which a model repeatedly chooses operations, a harness carries them out, and the results return to the model to inform its next choice.
- Context and context window — Context is the information available for a generation; the context window is the capacity available to hold it.
- distillation — Training a small, cheap model to imitate a big one's outputs, keeping much of the capability at a fraction of the cost.
Checking sign-in…
Loading comments…






