📍 Dubai | 🤖 I build agents that sell things | 🎨 Ex-Ogilvy, 17 years in advertising
Designer at Ogilvy, then creative director, then media buyer spending six figures a month of other people's money on direct response. Now I build the machines that do the whole job: agents that write, design, cut video, build offers, and argue with each other until something sells.
Selling was the hardest function to automate. It's the one I've automated.
- 🔎 eval-health - a deterministic linter for LLM eval results. Nine ways your eval lies to you, each caught by a rule, not a vibe. Born from a 40-config model tournament where the scoreboard lied before the models did. One Python file, no dependencies.
- 🎨 Hermes Skin Lab - design a terminal agent's whole personality in your browser. 30 skins, palette synth, ASCII art tools. Try it live
Building AI creative systems at Uponly and forward-deploying them into other people's businesses. Something new ships here every couple of weeks; this list grows.
- Distribution beats product.
- Intelligence is cheap. Deployment is the moat.
- Taste is the bottleneck, not the models.
- A small agent with good context beats a big model without it.
- The auction teaches you what design school doesn't.
Random facts
- Everything I write starts as a voice note on a walk.
- 17 years making ads taught me more about AI outputs than any benchmark.
- Three years in Dubai and still not used to July.