Les 7 mémoires d’un agent IA
A human doesn’t have one memory but several: what you’re doing right now, the facts you know, the experiences you’ve lived, the moves you’ve automated. An AI agent is built the same way. This grid sorts its memory into 7 categories, each stored in a different place. Knowing it helps you decide where to put what when you design an agent, instead of cramming everything into the conversation.
Strengths
- Gives a clear word to each kind of memory, so you can discuss it as a team instead of vaguely saying « memory »
- Separates two things people constantly confuse: what fits in the current conversation, and what must survive across sessions
- Helps you decide cleanly where to store each piece of info when you design an agent
Limitations
- It’s a reading grid, not a product: knowing the 7 boxes doesn’t tell you which one you actually need, and three are often enough
- Two categories (parametric, prospective) rarely serve day to day and can make you think you must build everything
Best for
- PMs, POs and designers who need to frame what an agent should remember without drowning the team in jargon
- Devs designing an agent who want to decide where to store what before writing a single line