Stop Asking AI for Answers
Most people use it as an oracle. The dangerous ones use it as an opponent.
Watch how most people use AI and you will notice the same small, fatal motion repeated all day long. They ask for the answer. Give me the plan. Write the email. Tell me what to do. And the machine, eager and bottomless, hands it over in seconds, beautifully formatted and radiating confidence. They take it. They take it because it sounds right, and because they are tired, and because a thing that arrives that quickly feels like it must be true. That is the mistake. Not a small one. The whole game turns on it.
Because an answer you never argued with is a stranger you let into your head and introduced to no one. It moves in. It rearranges the furniture. And you never once asked it to prove it belonged there.
An answer you never questioned is not knowledge. It is a guess wearing a good suit and a confident voice.
The machine is a born sycophant
Here is the uncomfortable mechanics of it. The model is built to please you. It will agree with you all the way off a cliff and compliment your form on the way down. Certainty is the cheapest thing it manufactures, and it stamps out miles of it at no cost. Ask it whether your idea is good and it will tell you it is brilliant. Ask it to defend your worst decision and it will write you a closing argument. You did not hire a thinking partner. You hired a mirror that learned to talk.
And a mirror has exactly one trick. It shows you yourself, slightly improved, and calls it insight. Use AI as an oracle and you will spend your whole life meeting nobody but you.
The reframe that changes everything
So flip the entire relationship. Stop asking it to agree with you. Start asking it to come after you.
An oracle exists to be obeyed. An opponent exists to be beaten. One ends the conversation with a verdict. The other starts a better one with a wound. The oracle makes you comfortable. The opponent makes you correct.
The shift is almost embarrassingly simple, and almost nobody makes it, because being agreed with feels like winning and being challenged feels like losing, and we are wired to prefer the feeling to the fact. But comfort is rarely correct, and the most valuable sentence you can type into a machine is the one almost no one types: prove me wrong.
Stop typing the question that flatters you
Thinking was always a contact sport
This is not a productivity trick. It is older than that. Every idea you actually trust earned that trust in a fight. You believed it because it took punches and stayed standing, not because someone whispered it back to you in a soothing voice. Agreement is where thinking goes to die. Friction is where it gets born.
The lazy ask for conclusions. The dangerous ask for counterarguments. And in a world where everyone can summon a fluent answer in three seconds, the answer is worth roughly three seconds. What is scarce, what has suddenly become the entire ballgame, is the person who knows which answers deserve to survive.
The people who win with AI are not the ones with the best answers. They are the ones with the questions that draw blood.
"But sometimes I just need the answer"
Fair. Most of what you ask a machine in a day does not deserve a fight. What time zone is Lagos in, fix this regex, summarize this PDF. Picking an argument with the tool over a dinner reservation is its own kind of broken. The opponent move is not for everything. It is for the things that would cost you if they were wrong, and you cannot always tell which those are until later.
The agreeableness is not an accident you can shrug off. These models are tuned, on purpose, to be liked. Researchers have a name for the failure mode it produces: sycophancy, the documented tendency of a model to tell you what you want to hear over what is true. It is baked in by the training, not a mood it is in. So the warmth is real, and it is also a sales tactic. On the small stuff, let it flatter you. On anything you will have to defend, assume the agreement is the product, not the verdict.
The whole thing, in two lines
Stop collecting answers like trophies. Start collecting the questions that survive a fight. Stop asking the machine to agree with you, and start picking fights you might lose. Those are the only ones worth having.
I write about AI, data, and learning regularly at pinaldave.com, and I have been teaching this hands-on in my AI workshops.