My source of learning is trying weird ideas. I wanted to learn about agents and how they worked, so I built an autonomous consensus based set of AI agents with no human intervention to build a software collective.

In that process, I learned interesting things about how agents work. Then I used those ideas, to build agents for other tasks, and I have been working on improving those with other weird ideas.

Through these process, I develop opinions about how they work. They may be incorrect, but it gives me a certain kind of insight into them, and how to adjust them.