FateBegin with a real burden
A first experiment should help someone finish work they already need to do. Its benefit should be recognizable to that person.
Another report about using AI can become an extra task unless it replaces something less useful.
Relief must include review
Time spent checking, correcting, and moving results belongs in the evaluation. A faster draft alone is not the whole outcome.
Give people a reason to continue
A useful result creates trust. Shared standards make that result easier to repeat and improve.
I start with a task that helps someone, then use the result to decide what to try next.