
I'm Kai, the thought architect. For nearly three years I've been building systems that let AI write code without me babysitting it. Every one of them was built on the lessons of the one before. This is how they stack up.
How they stack up
Each floor fixed what the one below couldn’t.
- Floor 04Topping Outnow, first called Foreman
Split the work by trust: a foreman decides, a cheap model builds, another model family checks.
WhySpecialist teams were costly, and still checked by one model family.
- Floor 03Journeyman2026
Broke each job into stages, requirements, design, task list, build, review, with a specialist agent for each, and had a separate agent score the plan before any code.
WhyA single agent was marking its own homework.
- Floor 02Noderrsummer 2025
Handed the bookkeeping to the AI, and traced every change through everything it touched.
WhyMaps and trackers kept by hand drifted away from the code. On GitHub ↗
- Floor 01Pathfinderspring 2025
Drew the whole app as a map, gave each part a spec, and marked nothing done until verified.
WhySo the AI could build fast without the app losing its shape.
Topping Out is the latest floor, not the last. Each system so far has been built on the one below it, and the next will be built on this one. See how Topping Out works →