Building with AI has changed where I spend my effort. I spend more time clarifying the job and deciding how to check it. That makes the work easier to direct and the result easier to judge.
These habits draw on architecture and delivery experience. They also help teams question assumptions that have become embedded in the current process.
Explain the outcome
State who needs the result, what they will do with it and what good looks like. Give the task a clear boundary. A request to “analyse this pack” is hard to assess. A request to identify unsupported assumptions, with source references, gives the reviewer something specific to check.
Provide the context that matters
Use approved information, relevant standards and examples of good work. Explain which sources take precedence when they disagree. Keep the material focused on the task so the reasoning and output can be checked against it.
Often the useful work is capturing judgement that an experienced colleague applies without having written it down.
Break the work at useful review points
Separate gathering evidence from drafting a recommendation when the task warrants it. Review the evidence before committing to the next step. Decide where a person needs to exercise judgement and what the workflow should do when information is missing.
Use the smallest set of steps that makes the work clear and reliable. Extra stages need to earn their place.
Leave a method the next person can use
Save the instructions, examples and checks that made the result useful. Explain where the method works and where it needs care. Give someone responsibility for maintaining it.
That is how an individual technique becomes a reusable team skill. Working alongside people on real tasks helps them develop the judgement to adapt the method themselves.