Most of what’s written about AI is written from the outside. This page is the inside view: what I actually run, build and depend on, in work and out of it. The essays linked below carry the detail; this is the map.
The work
I’m co-founder, COO and CTO at kash. Three of us built kash’s whole product. Mine was the fat middle, from the hooks the frontend calls down to indexing on-chain events, built with AI — not because AI typed faster, but because it deleted the coordination overhead that used to demand a team. What that doesn’t mean is fewer people: AI collapsed the build, not the operation. Running systems safely when money’s on the line still takes humans, runbooks and kill switches, and I build all three.
The practice
Claude Code is the daily instrument. A year of using it hard is distilled in In Clawd We Trust: the mental model, the environment, the phased workflow, the quality practices that compound. The short version — the constraint isn’t the tooling anymore. It’s judgement: knowing what to point twenty agents at, and what not to.
The personal stack
The same machinery runs my life outside work. A personal assistant that’s always on — running on my own hardware, reachable from the app I already text in, touching only what I let it. It’s wired into a second brain that doesn’t die after three months, which means the assistant answers from my own notes, in my own words, instead of guessing.
The view
I’ve been writing about where this goes since before it was fashionable: agents representing people and businesses directly, and the infrastructure they still don’t have — machine-speed rails instead of ones built for humans. The finance grounding matters here. It’s the hardest domain to ship technology into: regulated, real-time, unforgiving. Most other domains are simpler versions of the same problems.
If you’re putting AI to work somewhere the stakes are real, I’m open to the conversation — direct, or through DD Ventures for advisory.