Continual Learning for an AI Agent

AGENT

Goal: continuously improve
the agent from its experiences
without forgetting.

MODEL
LLM(s): weights
model selection

HARNESS
prompts · skills ·
tools · code ·
workflow

MEMORY
in-session state
persistent
knowledge

World
users · tools · data
policies

Agent logs / outputs