The work you automate first is how people learned.

Years ago, at a medical equipment delivery company, our warehouse repair supervisor had fixed so many oxygen concentrators that he could listen to one run for a few minutes and tell you what was wrong with it.

Nobody taught him that. The volume did.

When he left, the repair team was well trained. They could fix what came in. But no one ever had his ear, so they had to troubleshoot mechanically, working through the machine instead of listening to it.

The same company did something on purpose that I didn't fully appreciate until later. Every new hire, from managers to customer service reps, spent two or three days riding with a driver. You saw what the drivers dealt with at every stop.

I had also filled out paper tickets in the field and faxed them to the office. Later, when we won a national contract to replace another vendor's equipment with ours, I spent several weeks doing switch-outs in Kentucky and Massachusetts while the field teams scaled up. So when we automated the delivery tickets, I already knew many of the problems, because I had done the work.

That is the part of AI adoption nobody budgets for. The routine work is the first thing you automate, because it is repetitive and easy to measure. It is also where people built the base knowledge and judgment you rely on when something goes wrong. Automate it and the work still gets done. The learning stops.

So decide on purpose where the learning goes. Keep some of the routine work with people, rotate people through it, or put them where the work happens. Help new employees build that base knowledge of the business.

If AI does the first hundred repairs, where does your next diagnostician come from?