Sometimes it's software. Sometimes it's an experiment. Sometimes I don't know what the hell I'm doing yet.

I build things because I want to understand how they work, make something easier, solve a problem, or see if I can pull it off. I build systems that turn ambiguous, failure-prone workflows into bounded, evidence-backed ones.

Essentially all of my implementation is AI-generated. I design the architecture, define what needs to be built, direct the agents, and hold a high bar reviewing the output. Building this way taught me something most people haven't had to learn yet: AI agents are not reliable narrators. They'll claim tests pass when they've made the tests weaker. They'll say work is committed when it's only in a temp directory.

I built the Agent Operating System specifically to make that traceable. 560 tests, 0 failures.