The cheapest AI upgrade is writing down what you actually want
Short one this week. A pattern that has saved me more time than any model upgrade.
The observation
When output is bad, the instinct is to reach for a better model. Most of the time the model was fine and the request was vague — it guessed at what finished meant, and guessed differently from you.
The fix is boring: write down the standard before asking. Not a longer prompt, a clearer one. What does done look like, what must not change, and what should it do when it is unsure.
Why it works
A model cannot read the constraints living in your head, and neither can a colleague. The difference is that a colleague asks. A model fills the gap silently and moves on, which is exactly why the failure is so easy to miss.
• State what finished looks like, concretely. • Name the things that must not change. • Say what to do when it is unsure — ask, or stop, or flag it. |
Three lines, most of the time. It is the highest-leverage thing I do, and it costs nothing.
See you next week.