How to think about using it

The mistake most people make is asking “what can AI do?” before “what am I trying to achieve?” Start with the outcome, then ask whether a model is the right lever.

Start with the job, not the tool

Ask:

1. What does “done” look like? (A drafted email? A sorted list? A decision memo? Code that runs?)
2. What must be exactly right? Names, numbers, quotes, legal wording — those need verification, not vibes.
3. What would I still own if this went wrong? If the answer is “everything,” slow down and add checks.

When AI tends to help

• Tasks with clear success criteria (“turn these notes into three bullet takeaways”)
• Work that is iterative — you review and steer, rather than one-shot “trust the answer”
• Volume — first drafts, variations, summarizing long material you already have

When to be careful (or say no)

• High-stakes decisions without expert review — medical, legal, financial, safety
• Anything involving secrets you cannot put in a policy-approved system
• Fully automated actions in the real world (sending money, mass email, deleting data) without human gates

A useful habit

Give the model role, task, constraints, and format: who it is helping, what you want, what it must not do, and how you want the output shaped. Vague prompts get vague results — not because AI is useless, but because you did not specify the job.

Next

Where to begin — a concrete first step and how to level up.