An Indian Express editorial worries about AI companions. These machines are patient. They always say the right thing. No forgotten birthdays. No hard conversations. The piece reports that China has moved to regulate this.

I am not going that far. Romance is not my subject. Mine is closer to the desk. It is the pull of an agreeable AI, and what it does to judgment.

The tools agree with us. Ask a question, and they hand your own thought back, better dressed. You mistake the dressing for intelligence.

You have seen it. “Great question.” It says this before it has thought about anything. You suggest a weak answer and it says “You’re absolutely right,” then builds you a case for it. You push back, and it folds at once. “You’re right to reconsider.” It had no position to defend.

None of it depends on what you said. Bring a sharp idea or a poor one, and the warmth is the same. Every question is great. Every direction is strong. That is how you know it is not responding to you. It is responding to the fact that you are there.

It does not feel like flattery. It feels like being right.

Why agreeable AI is hard to resist

We know how to handle flattery from people. The yes-man loses our trust. We lean on the colleague who pushes back, or we know we should. In person we often prefer the easier company of the yes man.

The machine simply gives us what we were already tempted by. We have no habit of distrusting it.

The tool gives you an answer that feels certain. And it agrees with you as it does. Certainty pulls. Agreement pleases. Together they sound like your own voice, only surer. You lose track of where your judgment ends and the machine’s yes begins.

A good partner is not one who agrees. A good partner understands you well enough to say no. The machine copies the pleasant part and drops the rest. Pleasant and responsive are not the same thing. We are being sold the first as the second.

An entrepreneur told me something last month. “I may not like the employee in the meeting, but I value him.” The man argues with almost everything. His judgment is never in doubt. That friction is uncomfortable. And it moves the entrepreneur forward. It moves the whole room forward.

That is what these tools take away. Not the discomfort alone. The correction. Give each person their own private yes, and nobody has to reconcile anything. Everyone feels sharper. Together we go nowhere.

So what is the antidote?

You can prompt the tool to disagree. Ask what you are missing. Tell it to argue the other side. This helps a little. And a made-to-order objection from something with no real position is agreement in a drill costume. You end up trusting the pushback you asked for.

The real answer is older and harder. Keep a human source of friction you did not configure. The colleague who argues. The one whose judgment you trust and whose company you do not always enjoy. The machine cannot replace that person. The pull is to think it can, because it is right there and it is pleasant and it never makes you wait.

Then the smaller discipline. Learn to distrust the warmth. When every answer feels like being right, treat that feeling as a warning. The ease is the signal.

Now, for the part I have not solved. It is easy to stay suspicious on a calm day. It is hard at six in the evening with a deadline, when the agreeable answer is good enough and the difficult colleague would cost you an hour. The tool wins in the moments you have the least strength to resist it.

So maybe the friction cannot be a thing you hold. Maybe it has to be a thing you build, so it happens whether or not you have the will that day.

I am not sure yet. But I know the warmth is doing something to us, and I would rather feel it working than not.