What to hand to AI at work, and what to keep yourself

Most people pick an AI tool first and look for things to do with it. It works better the other way round: start with the tasks in your week and decide, one at a time, which ones are worth handing over.

Start with a list of real tasks

Write down ten things you did last week, from your calendar and your sent mail. Use real tasks, such as "weekly status report" or "reply to a supplier", not categories like "communication". You are sorting actual work, so the list has to be actual work.

Three questions for each task

Put each task through the same three questions:

• How often do I do it? A task you do every week repays the effort of learning to brief it. A one-off rarely does.
• What does a wrong answer cost? A slightly clumsy summary costs a minute. A wrong number in a client report costs trust.
• How fast can I check the result? If you can verify the output in a couple of minutes, handing it over is cheap. If checking takes as long as doing it, you have gained nothing.

Sort into three piles

Hand over: frequent, low cost if wrong, quick to check. Reformatting notes, a first pass at a routine email, turning your own bullet points into a draft.

Hand over, then check carefully: useful but the cost of a mistake is real. Research summaries, anything with figures, anything a client or your manager will read as your work.

Keep: decisions, anything you cannot verify, private or confidential material you should not paste into a tool, and the tone of a hard conversation. The judgement is the part that should stay yours.

Then test the first one

Pick one task from the first pile and time it with and without AI, counting the checking as part of the time. How you feel about the speed is a poor guide, so measure it. The free Unpuzzled Brief worksheet gives you the six fields to brief the tool properly, so the test is a fair one.