Agent Memory

Chapter 2. What the Agent Can Read and Write

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 2nd 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].

An agent that solves tasks on its own and an agent that draws on an organization’s institutional knowledge differ in how deeply each is wired into its surroundings. The first runs only on what you put in front of it. The second is able to reach for knowledge it was never handed, such as incident histories, runbooks, and other relevant documentation. In practice, the difference between these agents is not how much they can do on their own, which is a great deal, but whether and how an agent can stay connected to the context fast-moving organizations generate on a daily basis. Closing that gap methodically and effectively becomes most of the work of moving an agent into an organization, and it’s the focus of this chapter.

Chapter 1 demonstrated a memory-equipped agent that avoided a duplicate-charge bug because it could read an organization’s prior incident. However, that agent relied on a single, manually provided memory source. This chapter addresses the larger ...