Skip to main content

Lesson 1 of 7 · Why Intelligence Doesn't Compound

Stop prompting — start building memory

From Agents That Remember: Context Engineering & Memory: An interactive distillation of Anthropic engineer Lamis Mukta's talk on context engineering and agent memory — the shift from prompting a model to building “a system that prompts itself”. Progressive disclosure, versioning and hashing for concurrency, read-only vs read-write. Predict, drag, decide.


Start here

Stop prompting. Start building memory.

The idea that reframes agent building: you're not supposed to prompt the model — you're supposed to build a system that prompts itself. From Anthropic's Lamis Mukta, this is context engineering: how agents remember, so they get better every time. Predict, drag, and decide your way through it.
  • Speaker: Lamis Mukta · Anthropic
  • Time: ~15 min
  • Style: Interactive — predict & decide
A raw model is brilliant but forgetful. Give it a memory it manages itself, and its intelligence stops being a one-off — it starts to compound.

How agent memory grew up

One year, four eras — tap each to see the leap
  • 1 · CLAUDE.md
  • 2 · Memory tools
  • 3 · Skills
  • 4 · File-system memory

How this course works

Each lesson hands you something to do — predict an outcome, drag things into place, or sort what belongs where. You'll need to engage to continue. That's the point: you remember what you commit to.

Topics: ai, agents, context-engineering, memory, anthropic, interactive