Why I keep AI out of the final decision

I use an AI assistant every day, and I almost never let it make a decision. That is deliberate.

What I ask it for

Not “what should I do?”, but “give me your suggestions, with the pros and cons of each”. The assistant is good at laying out options and at the groundwork: drafts, comparisons, first versions. A person still chooses.

What happens when I push

I challenge the answer. Often I go further: I ask again, pressing the opposite view, sometimes a bit aggressively. The reply usually starts with “You are right.”

That sentence tells me what I need to know. An assistant that agrees with a position and then with its contrary is not judging. It is following me. If I let it decide, I would be deciding through a mirror.

This is why critical thinking has to stay with the person. The assistant can help me think. It cannot take the responsibility, and it cannot tell me when I am wrong unless I make it.

The two tests for what I do let it decide

Sometimes I let it take a decision. It must pass both tests:

1. No consequence for anyone else. If a colleague, a customer or a supplier would feel the result, it stays with a person.
2. Easy to reverse. If I can undo it tomorrow at no cost, a wrong answer teaches me something. If I cannot, it does not.

Choosing the order of my own reading list passes. Anything touching people, money or a commitment does not.

What this means for a team

If your team uses an assistant, agree on those two tests, and ask people to bring the options, not only the answer. A person who can explain why an option was rejected has done the thinking. One who forwards the assistant’s conclusion has not.

A note on speed

This approach is slower than accepting the first answer. It still saves the groundwork. The time saved is real; it simply stops short of the decision.

A question to keep

When did an assistant last agree with you too quickly, and did you notice?