I work with AI almost daily and I'm convinced of its potential. That's exactly why I want to spend one post on the other side of the ledger. The hype cycle is very good at telling us what AI can do. It's much quieter about the structural problems that come with it, and those problems are not hypothetical. Most of them are already visible if you're willing to look.

The Ones I Worry About Most

Start with quality. AI output is probabilistic: you get the most likely answer, which is not the same thing as the correct answer. The difference is easy to miss because the output looks so convincing, and that's precisely what makes it dangerous. A confident wrong answer in a legal document, a medical summary, or a financial report can do real damage before anyone notices. If we treat AI output as truth without verification, we'll eventually get burned, and probably in the place where we can least afford it.

The second one is subtler and, I think, worse: the more we delegate thinking to AI, the less we practice thinking ourselves. A junior developer who lets a model write all the code skips the struggle, and the struggle was where the learning happened. An analyst who accepts AI summaries without ever opening the source material slowly loses depth. Play that forward a few years and you arrive at an uncomfortable question: who is left to judge whether the AI is right? The expertise needed to verify the output erodes through the very act of relying on it.

And then there's dependency. Today's AI landscape is dominated by a handful of companies with the capital to train and run large models. I've written before about why I don't trust Big Tech with my personal data, and the same unease applies here at a larger scale. When your business depends on an API from a company whose interests may not align with yours, that's a risk you've chosen. When entire economies depend on infrastructure controlled by a few players, nobody chose it, and that's the kind of risk that tends to surprise people.

The Ones We Talk About Too Little

There are more. Entry-level knowledge work is under real pressure, because so much of it consists of exactly the tasks AI handles well: summarizing, drafting, screening. I don't believe those jobs vanish overnight, but I also don't hear a good answer to where the people who do them are supposed to go next.

Decisions are quietly shifting from people to systems, too. Hiring, content moderation, resource allocation. Efficiency is seductive, and every individual delegation seems reasonable, but the sum is a loss of human agency that nobody explicitly decided on.

And training and running these models costs enormous amounts of energy. I notice the environmental footprint almost never comes up when organizations discuss adopting AI. I don't have a neat answer for this one, but leaving it out of the conversation entirely feels wrong.

So What Now?

Not stopping, that much is clear. I use these tools every day and they're not going away. But there's a difference between using something and trusting it blindly. In practice that means checking output before it leaves the building, not letting junior people skip the learning curve that AI makes so easy to skip, thinking twice before wiring your business to a single provider, and deciding deliberately which decisions stay human.

AI amplifies whatever we point it at, including our blind spots. I'd rather we point it carefully.