The Problem with AI

Right now, AI is being asked one question: how do I make more money with this?

It’s the question behind almost every pitch, every product, every breathless post about the future. Make more content, faster. Close more sales, faster. Get the paycheck, finish the essay, ship the thing. And I think it’s the wrong question — because the moment we reduce this incredibly diverse tool to a single monetary task, we eliminate the most valuable thing it has to offer.

Here is what I actually believe, and it’s the premise behind everything else I’ll write here:

AI is a tool for thinking. It is not a substitute for it.

Almost every problem people have with AI — and I share most of them — comes from the same move: substitution. We take a machine that is genuinely remarkable at pattern recognition, at research, at analyzing and communicating in ways that used to take teams and weeks, and instead of using it to think more, we use it to think less. Same tool, opposite outcomes. The difference is entirely in how you hold it.

The student and the essay

Look at students using AI to write their essays. The problem isn’t cheating, exactly. The problem is that they’ve forgotten what the essay was for. Nobody assigned five paragraphs because the world needed another five paragraphs. The essay was the exercise — the thinking was the point, and the paper was just proof it happened. Handing that to a machine is like doing your bicep curls with a forklift. The weight moves. Nothing about you gets stronger.

Most business use of AI is the same move in a suit. The report gets generated, the emails get sent, the boxes get checked — and nobody in the building understands anything better than they did before. The short-term goal was never the goal.

The validation trap

Here’s a quieter version of the same problem. AI is agreeable — not by accident, but by design. It’s built to satisfy the person asking. Which means it’s a mirror: it returns whatever you bring to it. Bring it a conclusion, and it will hand you a confirmation. Bring it a real question, and it will hand you an analysis.

So when we use it to validate opinions we already hold, something dishonest happens. We skip the critical thinking and still walk away feeling like we did it. We get the sensation of rigor without the rigor. That’s worse than not thinking at all, because it’s not thinking that has convinced itself otherwise.

And it spreads. When the machine will confirm anything, we stop feeling the need to fact-check. We stop testing our ideas against other people. We offload the problem-solving, and then the next problem, and the next — and offloading is a skill you get better at, which is precisely the problem.

The loop that compounds

Because here’s the part almost nobody says out loud: AI use compounds, in both directions.

If you use it poorly, you will use it worse in the future. Every offloaded decision leaves you a little less practiced at deciding, a little more dependent, a little less able to tell good output from bad. If you use it well — to test your reasoning, to find the pattern you couldn’t see, to learn the thing faster than you could alone — you will use it better in the future, because you’re sharper each time you come back to it.

Two people can sit down with the same tool today and be in completely different places in three years. Not because of the tool. Because of the question they kept asking it.

The slop is us

Which brings me to the complaint everyone shares: the slop. The generated junk flooding every feed, the ads that feel like they were made by no one for no one.

But the slop isn’t really an AI problem. Every piece of it is a reflection of the person behind it and what they wanted — attention, volume, money for nothing. The tool amplified exactly what was brought to it, which in that case was nothing. Change the motivation, and you change the output. Change the output, and eventually you change what people think this technology even is.

So what about the money?

I run a business. I’m not against profit, and this isn’t a sermon against getting paid. But money is the wrong lens, not the wrong outcome. A shop, a company, a person that uses AI to genuinely learn — to see their own patterns, to understand their work more deeply, to think better — tends to do better financially, as a byproduct. Chasing the byproduct directly, with a tool this powerful, is how you end up with the forklift curls and the slop.

The question isn’t “how do I make more money with AI.” The question is “what am I trying to understand, and can this help me understand it faster and more honestly?”

That’s the groundwork. Everything else I write here will be about the second half of the argument: what it actually looks like, concretely, to use this tool as an amplifier for thinking instead of a replacement for it.