Tools · Writing Clear Instructions for AI

Watch what happens next. You type "write an email to the client about the delay," press enter, and get back something polite, generic, and not what you meant. The tool worked. Your instructions did not.

What separates a good instruction from a vague one

AI predicts from patterns. When your instruction is vague, you leave a gap, and the AI fills it with its most likely guess, the average of everything it has read. A clear specification closes that gap.

Specification thinking is describing a job precisely enough that an AI can carry it out and you can tell whether it succeeded. A good specification names three things: the goal (the outcome you want, stated as a result), the standard (what good looks like: length, audience, format, tone), and the limits (the constraints: what to leave out, what not to touch, what it must not invent).

Return to that client email. Instead of "write an email about the delay," you name the goal (reassure the client and hold the relationship), the standard (four short paragraphs, warm but professional, offering a revised timeline), and the limits (do not admit fault, and do not promise a date you cannot keep). Same tool, same model; the difference is entirely in what you asked for.

Firing off one-liners and hoping is the Aware habit; naming the goal, the standard, and the limits by reflex is the AI-Ready one. It is also your safeguard: an AI fills any gap confidently, even inventing a delivery date, so the limits keep a plausible guess from becoming a promise you never made.

Key Takeaway

A clear specification names three things, and the limits are what keep a confident guess from becoming a promise you never made.