For practical purposes an agent is learning if you put information in its context. So you can create a comprehensive architectural analysis and put it into context and suddenly the agent is familiar with the project. You can also delete information from context, which is a rather hard and imprecise process with humans.