This Is How AI Dependency Starts

More and more people say they can't work without AI anymore. Not that they prefer not to — they genuinely can't. The presentation won't start. The email won't move. The thinking doesn't come.

But it doesn't stop there. It gets smaller. A question comes up in a meeting — a simple one, the kind that used to just come out. It doesn't. Later, maybe. After checking with AI first.

Then smaller. A yes or no. A one-line answer. Something with a single right answer that you knew, once. You open the chat window anyway. Just to make sure.

That's not caution. That's dependency.

Most people assume that's just what happens when a tool becomes indispensable. It isn't. Dependency isn't caused by how useful AI is. It's caused by how you learned with it.

Ask a question, get an answer. Concepts, examples, depth — instantly. You cover in an hour what used to take days. Then a few days later, the same topic comes up. It's gone. Ask again. Gone again.

The reason is mechanical.

When AI corrects you immediately, your brain never has to retrieve it on its own. It just receives the correction and files it away. Smooth. Frictionless. And completely shallow.

Memory doesn't form from receiving — it forms from retrieving. The effort of pulling something back out of your head, even imperfectly, is what makes it stick. Every time you struggle to recall something, your brain is rebuilding the pathway to it. That struggle is the work. Remove the friction and you remove the memory.

A tutor optimizes for correction. The moment you're wrong, you're fixed. One pass through the material. Clean, efficient, and forgotten within the week.

Do that long enough, and the retrieval muscle stops working.

When I started studying law, I used AI as a classmate instead — someone starting from the same place, knowing nothing. We worked through lectures together, drafted mock complaints, argued about each other's claims. Lots of errors. Claims that didn't hold up.

But because my partner didn't know the answer either, getting it wrong played out differently. I didn't get corrected. I got: "What about this, though?" The wrong claim stayed in play — it became the next argument. I had to keep holding it in my head, turning it over, testing it against the next thing we read.

In my second year, I brought up something we'd gotten wrong as first-years. My partner disagreed. So I had to go back — not recall a correction, but rebuild the reasoning from scratch. Work through it again. Arrive at an answer myself.

Sometimes I was right. Sometimes I wasn't. But either way, my brain had touched that material three, four times — retrieving it, re-examining it, resolving it. A tutor would have corrected it once in year one and moved on. This version forced me to own it.

That's the mechanism. The wrong answer from year one survived into year two. When it came back, retrieval did the work that correction never could.

Wrong answers don't die. They become questions you have to answer yourself. That's not a side effect of learning this way.

That's the whole thing.