My daughter climbs into the warm car, frozen after an hour of riding lessons. "Dad, cars are so much easier than horses. They don't want to go anywhere. They just always listen." She had just spent an hour on a horse that kept turning around on the track and mostly did its own thing. And you know what's bizarre? Half an hour earlier, I stood by the fence with numb fingers, scribbling in my notebook: "horses + cars => AI. What we can learn from horse riding in the adoption of AI."

A Horse, Not a Car

She's right about cars, of course. Turn the key, press the pedal, the machine obeys. No questions asked. That's why we love them.

A horse is different. A horse has opinions. The one she rode that morning kept turning around on the track, and what struck me was that the instructor never once called it a bad horse. She asked my daughter: "What is the horse trying to tell you?"

That question has been stuck in my head ever since, because it's the question I keep wanting to ask in conversations about AI.

Most of us approach AI the way my daughter would prefer: like a car. Press a button, get a result, complain when the result is wrong. But these systems behave much more like that horse. They surprise you. They produce output that makes you mutter "why on earth did it do that?" And the answer is rarely simple disobedience. Usually the instructions were unclear, or the expectations were wrong, or we misunderstood how the thing actually works. Usually it's us.

What the Instructor Knew

So when an AI project runs into resistance, from the system or from the team that's supposed to use it, I try to remember the instructor's question. A model surfacing strange biases is telling you something about your data. A chatbot giving useless answers is telling you something about your documentation. And colleagues who quietly avoid the new tool are telling you something about how it was introduced. You can pull harder on the reins, but that has never once made a horse calmer.

Trust is the other thing you can't rush. A horse only gives it after consistent, patient interaction, and teams are no different. People won't trust a system that touches their work and their customers just because management says so. Honestly, they shouldn't. The slow route of small pilots, honest evaluation, and admitting what didn't work looks inefficient on paper. It's also the only route I've seen actually succeed.

And the best riders don't dominate their horses at all. They read them, adjust their posture, work with the animal's instincts instead of against them. Hold that image next to the dream of full automation, of removing humans from the loop entirely, and you can see the mismatch. The implementations that work tend to look like a partnership: AI bringing speed and pattern recognition, people bringing context and judgment.

Back at the Fence

My daughter did get that horse moving in the right direction, by the way. Not by pulling harder, but by sitting differently and giving clearer signals, until horse and rider understood each other for a moment.

Standing there with cold fingers, I wrote one more line in the notebook: we spent thousands of years learning to work with horses before we invented the car. With AI, we've barely left the stable. The complexity that frustrates us now is the same complexity that makes these systems worth the effort, so we'd better learn to ride.