The biggest takeaway for me was how a simple folder structure and a few instruction files can replace a lot of complex AI infrastructure.
Seeing how an agent can navigate folders and find the context it needs really made it click.
I've already started building my own AI Workspace using this approach. My goal is to maintain one source of truth that I can share across ChatGPT, Claude, and other AI tools.
Going forward, I want to keep improving that structure so I can switch between models, compare results, and avoid duplicating context.