An AI workflow request should explain the input, output, rules, examples, failure cases, and how the result will be reviewed.

AI projects get messy when the buyer only says "make it smart" without defining what good output means.

What details make AI tasks clearer?