How to think differently about AI
How to think differently about AI
A useful way to collaborate with AI is to separate the stages of the work: understand the task, explore options, choose an approach, produce a result and verify it.
These stages are a practice to try, not a scientifically established productivity formula. There is no universal percentage of work that should be assigned to a person or a model.
Name the audience, intended action, available evidence and limits. If the model would need to invent a fact, ask it to identify the gap instead.
For example: “Draft a service description using these business notes. Do not add prices, guarantees, qualifications or customer results that are not in the notes. List the missing facts separately.”
Ask for options and tradeoffs before committing to an approach. Record the choice you make and why. Do not let a polished draft quietly determine the business policy or product requirements.
Compare the output with the source material. For code, run a relevant test and inspect how failure is handled. For an unfamiliar factual claim, find reliable evidence. The model agreeing with its own answer is not independent verification.
Try the same bounded task once with a short prompt and once with an explicit brief and review criteria. Compare omissions, corrections and total time, including review. Keep the approach that helped with your actual task; do not infer a general performance guarantee from one attempt.