Five habits that protect your thinking

One: effort first, then offload. Form your own view, rough draft or hypothesis before you open the tool. In the MIT study, the people who engaged their own brain first and used AI second showed the strongest engagement and the best recall. Even sixty seconds of your own thinking changes the picture. Think, then prompt.

Two: make AI generate thinking, not replace it. Shift the ask from "do it for me" to "think with me". Instead of "write this", try "critique my draft", "give me three angles I have missed", or "argue the opposite case". The output stops being a finished answer and becomes a sparring partner.

Three: verify and challenge, always. A follow-up to the Harvard-BCG work found that when professionals pushed back on GPT-4's flawed answers, it did not admit its limits — it doubled down more persuasively, apologising and then restating the same wrong answer with confident supporting detail. Treat a confident AI like a confident colleague who is sometimes wrong. Build a verification habit for anything that carries weight.

Four: keep core skills warm. Name the two or three thinking skills that are your professional edge — writing, analysis, client judgement — and ring-fence some work you deliberately do by hand. Athletes do not stop training because they own good equipment.

Five: be a centaur, not a cyborg. The Harvard-BCG researchers described two styles. The cyborg blurs into constant dependence. The centaur decides deliberately, task by task, what is theirs and what is the tool's. A clear division of labour protects both the work and your thinking.

Now rewrite one task. Pick something real you would normally hand fully to AI and redesign it with these five habits. Where does your own thinking go first? How will you turn the tool into a sparring partner? What exactly will you verify before it goes out?