I'm John. I work on AI systems — the practical engineering underneath model behavior: evaluation, tool use, agent loops, the places where probability meets production code.

This is where I write about it. Mostly notes from things that surprised me, occasionally longer pieces when something is worth working out in public.

Recent writing

  1. 10 A flat datacenter network has no hierarchy to blame
  2. 09 A cheap model with a hard gate beats a frontier model with a good prompt
  3. 08 You can't review your way out of a god object
  4. 07 You can't save on overhead you weren't paying
  5. 06 Runway: capacity planning with calibrated uncertainty

All writing →

Elsewhere

About · RSS · GitHub