Skip to main content
Producing reliable AI agents is a manufacturing problem. That is this factory’s central bet. Agents today fail the way early production lines failed: defects pass silently downstream, quality gets inspected at the end instead of built in at each step, heroics substitute for standard work. The industry default is artisanal — one bespoke agent at a time, quality as good as the builder’s attention that day, nothing interchangeable, nothing certified. Craft production, with craft production’s ceiling. Toyota’s system is the best answer humanity has produced to exactly this shape of problem, and the mapping is direct:

​
Where the analogy strains — and what that teaches

Borrowing the discipline matters more than borrowing the vocabulary, and the places the analogy breaks are the places this factory’s own six theses live:
  • A Corolla’s quality is fixed at the gate; an agent’s quality is a distribution over future runs, and every input to that distribution drifts. So certificates expire and there are no finished goods.
  • A jammed press is unmissable; a dead agent process is inaudible. So the andon needs a second wire for silence.
  • Toyota’s cars do not build cars; this factory’s outputs staff the factory. So the recursion needs germline change control.
  • The expensive building is not the assembly line at all — assembly is nearly free. It is verification.
One strain matters most. Lean adopted as culture fails 80 to 95 percent of the time in human organizations. Agents make the fix possible for the first time: a human plant can only ask its workers to follow the standard; this plant compiles the standard into gates the work physically cannot skip. Toyota’s discipline plus doctrine that compiles — that is why Toyota, and why now.

​
Self-audit