The centerpiece skill

Asking precisely: the one skill that separates useful AI from disappointing AI

June 9, 2026 · 6 min read

When someone tells me "I tried AI and it was useless," I ask them to show me what they asked. Nine times out of ten, the problem isn't the AI. It's that the question would have stumped a human assistant too.

The vague question experiment

Imagine handing a capable new employee this instruction: "Clean up the customer list."

Clean up how? Remove duplicates — matched on what, email or name? Fix capitalization? Standardize phone formats? Delete the ones who haven't bought in a year, or flag them? Your new hire would come back with questions. AI mostly doesn't — it just guesses. And then you look at the guess and conclude AI doesn't work.

Now try this instead: "In this spreadsheet, find rows where the email address appears more than once. Keep the row with the most recent order date, delete the others, and give me a count of what was removed." Suddenly the machine is brilliant. Nothing about the AI changed. The question did.

Why this is the skill

Here's the framing I use for everything: the AI cooks, you fish. The AI can write the code, draft the text, do the mechanical work. What it cannot supply is what only you have — knowledge of your business, the rules of your work, and what "correct" looks like. Asking precisely is how that knowledge gets into the machine.

This is why I treat it as the centerpiece of all training, ahead of any particular tool. Tools change monthly. The skill of turning fuzzy intent into a precise, checkable request transfers to every tool that will ever exist — and it makes you better at delegating to humans, too.

What precision actually looks like

It's not about magic words or "prompt hacks." It's about supplying the same things a good contractor would need:

• Context: what the data or situation actually is. "A spreadsheet where column A is customer email and column F is the last order date" beats "my customer list."
• The rules: every decision you'd make by hand, written down. If you can't articulate a rule, that's a sign this part needs your judgment, not automation.
• What "done" looks like: describe the output you want concretely enough that you could check it. "A count of removed rows" gives you a way to verify.
• Small steps: one precise request, check the result, then the next. The people who get the most from AI work in short verified steps, not one giant wish.

A precise question is really a description of your own work, made explicit. That's why asking precisely makes you better at your job even before any AI is involved.

The honest caveat

Precision raises the ceiling; it doesn't remove it. A perfectly-asked question can still hit a genuine AI limit — and knowing those limits is its own skill (I keep an honest list here). But you can't even see the real limits until vague asking stops generating fake ones.