Agent Memory

Chapter 3. Choosing What Becomes Memory

A note for Early Release readers

With Early Release ebooks, you get books in their earliest form—the author’s raw and unedited content as they write—so you can take advantage of these technologies long before the official release of these titles.

This will be the 3rd chapter of the final book. The GitHub repo will be made active later on.

If you’d like to be actively involved in reviewing and commenting on this draft, please reach out to the editor at [e-mail].

Organizations don’t only accumulate critical knowledge in documents. Interactions between users and agents are also where corrections, decisions, and shared understanding emerge. Recall Example 1-3, where organizational memory changed how an agent investigated the vendor timeout bug. But so far the system of our Very Important Company running example only handles bringing existing memory into the conversation. It doesn’t yet decide whether something learned during that conversation should persist as separately managed memory.

That missing decision becomes a scaling problem quickly. In a large enterprise, even one daily exchange per employee could produce hundreds of thousands or millions of interactions. No one person could inspect every conversation and decide which information deserves durable retention. Selection therefore has to become a system responsibility. At that scale, memory candidate identification and evaluation must be automatable, while decisions that ...