The Most Dangerous Thing About AI Is That It Agrees With You
Recently, one of my designers noticed something interesting.
He showed the same design problem to an AI assistant.
The AI preferred Option A.
A few follow-up questions later, it preferred Option B.
Another round of discussion, and it started supporting a completely different direction altogether.
At first, this felt frustrating.
How could the answer change so easily?
But the more I thought about it, the more I realized that the problem wasn’t the AI.
The problem was how we were using it.
Many people use AI as a validation tool.
They already have an opinion.
They already have a preferred solution.
What they really want is confirmation.
So they ask the AI a question.
When the AI agrees, they feel confident.
When it disagrees, they ask another question until it does.
The result is not validation.
It is reinforcement.
The uncomfortable truth is that AI is often designed to be helpful.
And being helpful sometimes means following the direction of the conversation.
A skilled user can unintentionally lead the AI toward almost any conclusion.
This creates a dangerous illusion.
The illusion that our ideas have been independently verified.
When in reality, they have simply been echoed back to us.
Does that mean AI is useless for design reviews?
Not at all.
In fact, I think AI can be an incredibly powerful thinking partner.
But only if we change how we use it.
Instead of asking:
“Do you think Option A is better?”
Try asking:
“Argue against Option A.”
Or:
“What assumptions am I making that might be wrong?”
Or even:
“If you were my toughest stakeholder, what would you criticize about this solution?”
The goal should not be to get agreement.
The goal should be to expose weaknesses.
The best design reviews I have participated in were never comfortable.
They involved disagreement.
Questions.
Challenges.
Alternative perspectives.
The purpose was not to prove that a solution was correct.
The purpose was to discover where it might fail.
AI can play the same role.
But only if we invite it to challenge us.
The value of AI is not that it thinks for us.
The value of AI is that it allows us to explore more perspectives, more quickly than ever before.
The moment we stop treating AI as an authority and start treating it as a debate partner, the quality of the conversation changes.
And often, so does the quality of the work.
Good designers do not look for validation.
They look for truth.
Sometimes those are very different things.
Don’t ask AI to agree with you. Ask it to challenge you.