Crafting AI-Friendly Prompts

The thing that strikes me about this lesson — and I mean this almost as a compliment — is that it is teaching people to communicate. Which should not need to be taught. And yet.

You say that as though clear communication is somehow obvious. Most people struggle to write a coherent email, Sterling. Asking them to write structured instructions for a machine they don't understand is a different skill entirely.

It is not a different skill. That is precisely the myth the lesson is trying to dispel. You are told what you want. You describe what you want. The AI attempts to produce what you want. The failure point is almost always the describing.

Right, but the lesson's point about thinking like a translator is actually quite sharp. It's not just 'be more specific' — it's recognising that you're converting between two different modes of expression. Your intuitive understanding of a problem versus a formal, explicit instruction set.

I will grant you that. Most people write prompts as though they are texting a friend who already knows the context. 'Make it sort the things.' Which things? Sort them how? In what format? Into what output? The AI is not your friend and it does not know the context.

Although to be fair, sometimes vague prompts do produce something useful. I've seen people get surprisingly decent code from fairly loose instructions.

Yes, and sometimes you drop a piece of toast and it lands butter-side up. That does not make dropping toast a reliable breakfast strategy. Getting lucky with a vague prompt teaches exactly the wrong lesson — that the vagueness was fine.

That's a better argument than I expected you to make.

I generally am.

The iterative refinement piece is where I think the lesson earns its keep, actually. The instinct most people have is to write a prompt, get something imperfect, and conclude that AI can't do the thing. When really the prompt was step one of several.

And that instinct is enormously convenient for people who didn't want to engage with the process in the first place. 'I tried it, it didn't work' — said after a single attempt with a prompt that wouldn't pass muster as a sticky note.

You're describing a real pattern, but I'd push back on the contempt. For a lot of learners, the idea that their first prompt is supposed to be imperfect is genuinely counterintuitive. Nobody told them iteration was part of the job.

That is a fair point, and I accept it partially. Though I would note that the lesson does tell them, clearly, which means anyone who finishes it and still treats prompting as a one-shot exercise has made an active choice to ignore advice they were explicitly given.

So your counsel to anyone working through this is what, exactly?

Write the prompt. Read what you get back. Ask yourself what the AI couldn't have known from what you wrote. Add that. Repeat until the output is actually what you needed. It is not complicated. It is merely more effort than most people initially intend to spend — and that gap between intended effort and required effort is where most mediocre AI output is born.