The prompt is the new spec, and most specs are bad

The vague ask, answered exactly

Someone showed me an AI result they were frustrated with. The output was wrong, they said, the tool was clearly useless. So I read their prompt. It was three lines, vague about the audience, silent on the goal, unclear on a single constraint. The AI had answered exactly what was asked, cheerfully and completely. The problem was that what was asked wasn’t what they meant.

AI turned everyone into a spec writer overnight. It did not, unfortunately, make anyone good at it.

Clarity was always the hard part

A prompt is a specification. And writing a clear specification, saying precisely what you want, for whom, within what limits, and what good looks like, was always the genuinely difficult skill in building anything. It was hard when the recipient was a developer, hard when it was a designer, and it’s exactly as hard now that the recipient is a model. The tool that turns the spec into output changed. The discipline of knowing what to ask for didn’t move an inch.

• A vague prompt gets a vague answer, delivered with total, misleading confidence.
• The real work is in the thinking that happens before the prompt, not the typing of it.
• If you can’t describe what good looks like, no tool on earth can produce it for you. It’ll produce something, and that’s worse.

AI will build exactly what you ask for. The skill has always been asking for the right thing.

A good prompt looks like a good brief

When I get a strong result from AI, the prompt almost always reads like a brief I would be happy to hand a freelancer. It names the audience. It states the goal in one sentence. It lists the constraints, what to avoid, what matters most, what good and bad look like. It gives an example. None of that’s a prompt-engineering trick. It’s just clear thinking, written down, which is the thing most projects are quietly missing well before any AI enters the picture.

This is also why the same skill protects you downstream. A model is a fluent junior that never says it’s unsure, which is exactly the trap I described in AI is a great junior and a dangerous senior. A sharp spec is how you keep its confident wrongness from becoming your problem, and it’s a big part of how I actually work day to day, which I wrote up in how I actually build with AI.

What this means for your work

Treat the prompt like the spec it is. Before you write a word to the machine, get clear on the audience, the goal, the constraints, and what good looks like, the same questions a good consultant would ask you before touching anything. The people getting the most out of these tools aren’t the fastest typers or the cleverest prompt-hackers. They’re the clearest thinkers, which is exactly what they were before the tools arrived. If you want a second set of eyes on how AI fits into your own process, I’m glad to talk it through.