AI is useful enough that the temptation is to use it for everything. That is not the same as using it well.
Use AI when the task is draftable
Code drafts, documentation outlines, alternative approaches, test cases and repetitive transformations are often good candidates.
Be careful when the task depends on business context
Metric definitions, policy interpretation, sensitive decisions and ambiguous stakeholder requests require stronger human ownership because the missing context may matter more than the syntax.
Use a simple four-question test
- Can I describe the expected output?
- Can I independently test the result?
- Are the business rules explicit?
- What happens if the answer is wrong?
Then choose the level of AI assistance
AI can brainstorm, draft, review or explain. You do not have to hand over the entire task just because the model can produce an answer.
That decision framework is central to my first ebook on practical AI-assisted analysis.