AI is a Drug

You’ve seen the post. It pops up on X every few weeks, right on schedule.

“Is it just me, or are AI coding models getting worse?”

Worse? Really?

Are AI Models Really Getting Worse?

Let’s get some perspective. Three years ago, we were clapping like seals if a model could finish a for loop without hallucinating a new programming language. Today, that same tool can scaffold an entire web app, write documentation, and patiently explain its own logic.

That’s not regression. That’s your dopamine tolerance kicking in.

The Dopamine Trap of AI Dependence

AI is a drug.

And like any powerful substance, it can open your mind or quietly take it over.

If you’re using it to replace critical thinking, you’ll always be chasing a stronger hit. Remember the first time you used AI? It felt like pure magic. Now, that initial high is gone. You’re scanning headlines for the next model, the next update, the “smarter” thing, because the old one just doesn’t hit like it used to.

But the model didn’t get dumber. Your relationship with it did.

That’s dependency. The quiet kind that trades curiosity for convenience. You stop challenging yourself. You stop learning. You stop owning your work. And when the API goes down, that low-key panic hits: wait… do I even remember how to solve this myself?

Using AI as a Teacher, Not a Crutch

But here’s the flip side: AI can also be a really good drug – the mind-expanding kind.

Treat it like the world’s most patient mentor, on call 24/7. It’s the teacher who never gets annoyed.

You can make it explain recursion five different ways – as a story, as a metaphor, with code examples, and then as a step-by-step trace – until that light bulb finally turns on. It doesn’t roll its eyes. It doesn’t sigh and say, “We went over this.” It just meets you exactly where you are.

That’s the real high. The learning hit.

The moment a concept that’s been confusing you for weeks suddenly clicks because you had a tireless guide to walk you through it one more time.

That’s not addiction. That’s acceleration.

The best developers I know aren’t the ones getting AI to write the most code. They’re the ones learning the fastest because they treat AI as a sparring partner. They don’t just copy-paste the answer, they challenge it.

“Why’d you pick that method?”

“Could this be cleaner?”

They argue with it, poke holes in it, and use that feedback loop to sharpen their own instincts.

So when someone says, “AI models have plateaued,” I have to wonder if it’s the models… or if it’s us.

The Real High: Understanding, Not Output

Maybe we just built up a tolerance for wonder. We got lazy. We forgot that the point was never to offload our thinking—it was to supercharge it.

AI hasn’t stopped getting better. The novelty just wore off.

The real breakthrough is remembering how to use the incredible ones we already have—as teachers, as collaborators, and as mirrors for our own curiosity.So yeah, AI is a drug. If you’re just chasing a stronger high, you’ll always be disappointed.

But if you’re chasing understanding, you’ve already got a lifetime supply.