The shapes of agent memory: how AI remembers things

An agent is just a language model with a tool belt — it can search, click, run code. But it forgets everything between turns. Memory fixes that. The post lays out four shapes: semantic memory for facts you want to keep long-term, episodic memory for what actually happened, working memory for the current context window, and procedural memory for how to do things. Each has a different storage strategy and a different latency tradeoff.

The real work is in the retrieval layer. You can't just dump everything into a vector store and hope — the post walks through chunking strategies, metadata filters, and the difference between embedding everything versus storing raw text for exact matches. The trickiest part is deciding what to keep. Too little memory and the agent repeats itself. Too much and it bogs down with outdated context. The author's take: start with episodic memory and a simple retrieval, then layer semantic and procedural as the agent gets more complex.

Why this matters for us: our abuelos and abuelitas are already being served by agents that forget their names, their allergies, their preferences — and this post shows the engineering that could actually make those systems reliable for real people instead of just another demo.