Two Buckets. Clean Record. Both Directions.
Two years of production AI experiments produced a clean finding. Every experiment designed to replace a person failed. Every experiment designed to enhance a person won. The record is clean in both directions.
Two years ago we started running AI experiments at scale inside a real retail operation. Not pilots. Not demos. Production deployments, with real customers, real workflows, and real consequences if they failed.
We ran a lot of them. Early on I noticed the results were falling into two distinct buckets.
Bucket one: experiments designed to replace a person with AI. Automate the judgment. Remove the human from the loop entirely. Every single one failed. Not most of them. All of them.
Bucket two: experiments designed to give a talented person more capability. Keep the human in the loop. Use AI to expand what they could see, decide, and execute. Every single one worked. And worked bigger than we expected.
That pattern was consistent enough that it stopped feeling like a finding and started feeling like a rule.
AI is not a replacement for human judgment. It is an amplifier of it.
The organizations that understand that distinction early will compound their advantage. The ones still trying to build the robot will keep wondering why it isn't working.
The record on this is clean. Both directions.