Zero to one is the part I enjoy most. The problem, interface, and code can still change together. Early decisions matter, but changing your mind is still cheap.

Build the small version first. It reaches reality faster, and reality is good at finding requirements that were invented in a meeting.

I like open source because the work stays inspectable. People can fix it, adapt it, or use it for something the original plan missed.

Agents, local-first apps, and developer tools look like separate interests. The common bit is simple: computers should handle more of the repeated work without hiding the important decisions.

Not every experiment needs a business model. Sometimes understanding how the system behaves is a good enough result.