Most bad systems are not broken. They run as designed, hit their targets and pass their audits —
for someone. The useful question about any system is not whether it works but who it works for,
and the answer is rarely everyone standing inside it. Take a support queue measured in response
time: it serves the dashboard, not the person answering and not the person waiting.
A better design follows two rules. Build on what people are good at instead of forcing them into
work they are bad at — people are superb at conversation and poor at being a waiting line. And
inform rather than control: a warning that something looks wrong is information, while a tool that
polices how someone works is control wearing quality's clothes.
Neither rule can be settled from outside. A written plan is a snapshot of what someone understood
on a Tuesday, and by the time the work is finished the world has moved and the snapshot is a
fossil. So we build the smallest honest thing, put it in front of a real person, and let what
comes back change the plan — steering by what we notice, rather than by what we predicted.
We hold the same standard for ourselves. The company is a system too, we are the people who
depend on it, and it is meant to get better at getting better.