Definition
Plain language
A memory system that builds up a skill library for an agent by recording successful workflows.
As stated in the literature
Agent Workflow Memory, a procedural-skill-library approach that abstracts successful trajectories into reusable workflows; learns only from wins, contrasted with Auto-Dreamer which also learns from failures.
Why it matters: Saving successful procedures turns one-off wins into a growing toolkit, but the design choice to learn only from wins means failures don't contribute to improvement.
For example, after successfully booking a flight, AWM might extract the click-sequence as a reusable 'book flight' skill the agent can invoke later.
Heard on the show
“That's huge, and it's the second case study — the head-to-head against AWM on the find-entity tasks.”Episode 064 — When Agent Memory Stops Being a Database and Starts Being a Skill