There is a cost in making software. A hidden cost that a lot of projects fail to evaluate. And worse, this won’t get better, it will be far worse before anything better comes out…

Let’s start with what the tool actually is. An LLM is a language model, a very fancy autocomplete, a mechanical parrot trained on a mountain of text, images and code it never asked permission to use. It predicts the next token, and sometimes it predicts a useful one. That’s the whole trick. The problem is never that it predicts, the problem is how we decided to sell the prediction. AI is a tool, and the only thing that matters is direction, usage, intention. Use it to proofread an email, to summarize a wall of text, to find a pattern in a pile of markdown, and it’s fine, a glorified spellchecker with a big vocabulary. Use it as the pilot and you get slop.

But the tool is so useful!

Yes, and a hammer is useful until you use it to hit your own thumb. Usefulness was never the question, the intention behind it was.

The Theos of the internet don’t want to hear that. For the third year in a row they announce that the coders are gone in six months, that knowledge work is over, that the machine will replace the room. It’s the same lie every cycle, and it always comes with a receipt. The pneumatic nailer didn’t delete the construction workers, and the tractor didn’t delete the farmer. A tool changes the work, it doesn’t erase the person holding it. But that’s a boring headline, so we get the other one, the one where the tool becomes the worker and the worker becomes the cost.

So no, “AI” isn’t the problem. The problem is the intention we pour into it, the slop we accept as output, and the fact that [they] keep building the tool around the demo and the demo around the tool…