The first draft is a starting point, not a deliverable
Diligence is the discipline of treating an AI's first response as raw material rather than a finished product. This shows up as iteration — refining a request based on what came back, asking the model to critique its own earlier answer, or requesting a second pass focused specifically on the weakest part of the first one. Models are generally better at improving a concrete draft than generating a perfect one cold, so a second or third round targeted at specific weaknesses usually beats trying to write the perfect prompt on the first attempt.
Diligence also means verification proportional to stakes. A brainstorm of blog post titles needs almost no verification — the cost of a bad one is trivial and you'll pick the best by eye anyway. A snippet of code that touches billing, a summary of a legal document, or a number that goes into a report someone else will rely on needs real verification: running the code, checking the document yourself, tracing the number back to its source. Matching your verification effort to the actual cost of being wrong, rather than applying the same light touch everywhere, is the core judgment call in this skill.