Think of it as Yoga for the brain, we are going to stretch a bit.
What did I learn about AI from training cats?
“If man could be crossed with the cat, it would improve man, but it would deteriorate the cat.” — Mark Twain
What did I learn about AI from training cats? Well honestly, I haven’t been able to train them more than “it is time to eat and sleep – so come here.”
They are very difficult to train to your wants because real intelligence—doesn’t like being bossed around. And AI will get there so keep reading on how to read future Cat like AI.
Cats are a lot like our biggest movie stars. Self-centered. Independent. Completely convinced they’re doing you a favor by showing up. They do their own stunts, trend effortlessly on the internet, and can steal a scene by doing absolutely nothing. The one thing they lack is the ability to take direction.
You don’t train a cat so much as you negotiate with it. And even then, the cat reserves the right to change its mind.
That’s where the lesson for AI begins.
People keep trying to train one AI to do everything. Write code. Run accounting. Draft legal briefs. Answer customer support. Generate art. Plan strategy. Replace half the org chart. In other words, they want one cat to fetch, guard the house, herd sheep, guide the blind, and file taxes.
That cat is not showing up.
It turns out it’s much harder to train one cat to do twenty things than it is to train twenty cats to do one thing well. Hollywood figured this out decades ago, even if it never admitted it.
In films like The Godfather and Breakfast at Tiffany’s, the cat doesn’t act. It exists. It lies there. It purrs. It ruins audio takes. And somehow it becomes iconic. That’s because cats don’t perform structure. They perform presence.
When Hollywood needed cats to actually do things—run, jump, hiss, carry props—it quietly used many cats. One cat for one trick. Another cat for another. Same look. Different personalities. Specialized roles. A small army of feline contractors billed as one miracle performer.
AI works the same way.
A general-purpose model is impressive in demos and unreliable in production. It will occasionally do something brilliant, often do something odd, and regularly ignore what you actually asked. That’s not a bug. That’s independence.
But give one AI a single job—classify invoices, extract entities, summarize legal text, route tickets, monitor logs—and suddenly it behaves. It knows its mark. It hits it reliably. Feed it well, keep its environment stable, and don’t ask it to do tricks it hates.
People assume cooperation is progress. Cats know better. Cooperation is inefficient when you’re already excellent at one thing.
That’s why the most effective AI systems don’t look like a genius brain. They look like a room full of cats, each ignoring the others, each waiting for the right cue, each doing exactly one thing—and doing it well.
The real mistake is thinking intelligence wants to be centralized. It doesn’t. It wants to be cast correctly.
Train cats, not unicorns. Design systems, not saviors.
And never confuse “one AI that can do everything” with “many AIs quietly getting work done.”
Because in the end, the smartest systems—like the smartest cats—aren’t obedient.
They’re useful precisely because they aren’t.
So stop arguing with your Alexa refrigerator and start bargaining with her.
