whoami
I’m a product manager. The day job is operational tooling — measuring whether what we ship actually closes the gap between dashboard metrics and the customer experience.
Outside the day job, I run a workshop. OpenClaw is the substrate — a 22-agent system on bare-metal hardware that I use to probe how Claude actually works: where orchestration breaks, where integration barriers show up, how different agent harnesses behave under load. Its Press Workshop side has shipped two commercial novels. Its editorial desk ships this site. A paper-mode options engine runs on the same infrastructure.
The question I’m working on: where does an LLM add real edge if you’re not throwing large-scale compute at the problem? There are obvious uses: 10-K extraction, sentiment, long-term hypothesis evaluation, no-code scaffolding for high-frequency training setups. But the gap between “useful” and “edge” is where most of the actual work lives. That’s what I’m chasing.