How I prompt: say exactly what you mean

I learned this learning a language

I spent two years learning Chinese from almost nothing. It drilled one lesson into me: language only works when it’s exact. Be vague and you’re misunderstood, by people and by models. Prompting is the same skill. It’s not magic words; it’s clear thinking, said clearly.

Describe the destination, not the directions

I don’t tell the AI how to do something. I tell it exactly what “done” looks like, how it behaves, how it feels, what happens at the edges, and let it find the path. Describe the outcome vividly enough and the steps take care of themselves.

When it’s wrong, name what’s wrong

“Fix it” is a wasted prompt. I say what’s off and what right looks like instead. Every correction is just a sharper description than the last one. That’s the whole loop, each round, the language gets tighter and the product gets closer.

Build through conversation

Nothing good comes out in one shot. The product gets made in rounds, and prompting is a dialogue that converges on the thing in your head. Treat it like you’re describing something that already exists and you’re just helping someone else see it.

Vague in, vague out

When the output is mushy, my input was mushy. The fix is almost always on my side. Before I blame the model, I ask: did I actually say what I meant? Usually I didn’t, so I say it again, exactly.