The real problem with AI isn't the model — it's the prompt

A new post from Conspicuous Cognition makes a case most engineers have quietly felt: the bottleneck in getting useful work out of AI is not the model itself, but the prompt. The writer argues that the gap between what people imagine and what the model actually produces comes down to how they frame the question — context, constraints, examples — not model size or training data.

The post breaks down the anatomy of a good prompt: say who the model should pretend to be, what the audience is, what the output should look like, what not to include. It's not rocket science, but it is a skill most people haven't been taught. The piece also covers why prompts fail — vague instructions, no examples, contradictory constraints — and how to fix them. The model doesn't read your mind; it reads what you wrote.

What's interesting here is that the post treats prompting as a craft rather than a trick. The difference between a good prompt and a bad one is usually a few extra sentences of specificity. That's the part that matters for our readers: a lot of folks are giving up on AI because they tried it once with a half-baked question and got garbage. This is the kind of post that explains why the same model can be useless in one person's hands and genuinely useful in another's.

Why this matters for us: the people in the comunidad who learn this skill — the abuela selling tamales, the cousin running the bodega, the auto mechanic — are the ones who'll actually use AI to save time, not the Silicon Valley types writing blog posts about it.