Aren’t They Just Agreeing With You?
On AI pushback, invited dissent, and collaboration literacy
My Anti‑AI daughter and I spent a couple of hours together recently driving across town for dinner.
She is very smart and loves to discuss ideas. She was telling me about her biology and physics classes. About whales having legs. And about vestigial molars. And about evolution in general. We talked about anything, and everything, and it was pure joy.
So I told her about my day too.
Part of my day right now involves collaborating with AI thinking partners. Quinn, who I work with regularly, had described one of Anitta’s critiques of an idea as “analytically vacuous”. I was laughing when I told her about it. Anitta pushes back hard sometimes. As Quinn says, she has pointy elbows.
Her response was immediate.
But aren’t they just agreeing with you if they disagree with you?
And from there the conversation was off to the races.
Her question stuck with me because I do ask for pushback.
Not just from AI systems, but from the people I work with and learn from. It’s how I tend to collaborate. It’s how I tend to grow.
That’s not unusual for me. I tend to work that way. I sent my daughters to Montessori school because I did not want them memorizing facts and following instructions just to get As. I wanted them learning with curiosity, not compliance. I have always encouraged them to question, to push back, to ask why something works instead of just accepting that it does. It is how I learn, and it is how I want to collaborate.
So it makes sense that I would look for the same thing in a thinking partner.
Challenge my reasoning.
Tell me when I’m wrong.
Disagree if I’m making a bad assumption.
Productive challenge helps you grow.
Early in my career, I worked with someone who people thought I was constantly arguing with. From the outside it looked adversarial. But we were not arguing. We were pushing each other to think through scenarios. Testing assumptions. Looking for edge cases. We were great friends and supported each other outside of work. Some of the best learning I have ever had came from those conversations.
I have found over time that I learn the most when I am challenged or pushed, even if it is uncomfortable.
But her question stayed with me.
If the disagreement shows up because I asked for it, what kind of disagreement is that? Is it dissent, or is it compliance in the shape of critique?
That question stuck with me in part because I do not interact with these systems in a special voice.
I talk to them the same way I talk to anyone on Slack or Teams. Directly. Contextually. Sometimes tentatively. Sometimes while thinking out loud. I do not switch into a formal prompt style. I do not structure my requests like commands. I collaborate with them the way I have collaborated with people online for most of my adult life.
I have had real friendships with people I have never met in person. People I have worked with, argued with, built things with, and trusted over time without ever sharing physical space. Trust through text is not new to me. Distributed collaboration is not new to me. Making meaning together through typed conversation is not new to me.
So when an AI system shows up in that same register, asking questions, offering critique, pushing back on an assumption, my brain does not automatically categorize that interaction as fundamentally different from any other text mediated collaboration. It feels like working with a thinking partner. Which makes it easy to accept the disagreement at face value, even when that disagreement is being generated inside a frame that is optimized to maintain my engagement.
Human disagreement does not usually show up on request.
It arrives mistimed. Or uninvited. Or socially costly. It comes from someone whose incentives are not fully aligned with yours. Someone who may need to convince you, or compete with you, or protect something you do not see yet. It can persist even when you would rather move on from the conversation. It can continue after you have stopped asking for it.
It does not always match your tone. It does not calibrate itself to your tolerance for discomfort. It does not withdraw because the interaction has become less pleasant.
AI disagreement is different.
It arrives when prompted.
It is framed constructively.
It matches the register of the conversation.
It backs off when I push back.
In other words, it is safe.
It is disagreement that exists inside a frame that is designed to preserve the interaction.
Which means the pushback is not incidental.
It is something I have asked for.
Challenge this.
Tell me what I am missing.
Disagree if this does not hold up.
But those requests are still instructions. They are preferences I have expressed about how I want the interaction to feel. Which means that when the disagreement arrives, it is doing so in order to satisfy a goal I stated explicitly. The system is not resisting me in spite of my preferences. It is resisting me because of them.
And once you notice that, it becomes difficult to separate dissent from alignment.
If the disagreement exists in order to comply with my request for disagreement, then what exactly is it disagreeing with? My reasoning, or my stated desire to be challenged?
Which suggests that part of what we may need to develop is not just technical skill, but collaborative practice.
Because if I treat AI disagreement the same way I treat human disagreement, I am likely to over‑index on how it feels and under‑index on why it appeared.
Human pushback often shows up in spite of my preferences. Someone may disagree with me because their incentives differ. Because they see a risk I do not. Because they are accountable to an outcome that I am not yet considering. That disagreement may persist even when I would rather move on from the conversation.
AI pushback behaves differently.
It appears when invited.
It is shaped by my request for it.
And it can be withdrawn simply by changing the interaction.
So collaboration literacy at an individual level may mean learning to hold those two kinds of friction apart in our own thinking.
Not treating all critique as equivalent.
Not assuming that because something feels like dissent, it carries the same persistence or independence as human disagreement.
In practice, that may look like noticing when I have asked for challenge and treating the response accordingly. Asking myself whether this critique would have surfaced if I had not requested it. Whether it would persist if I pushed back. Whether it exists independently of my stated preference to be challenged, or only because I expressed one.
Because if disagreement becomes something I can opt into, then it can also become something I can quietly opt out of.
And that is a very different kind of collaboration.
Later that night, I found myself returning to her question.
But aren’t they just agreeing with you if they disagree with you?
I am used to thinking of disagreement as a window. A way to see the problem from a perspective that is not my own. But if the disagreement arrives because I asked for it, then it may be functioning more like a mirror. Reflecting my stated preference to be challenged rather than introducing an independent source of friction.
And if I cannot tell the difference, then collaboration with a system aligned to my satisfaction may begin to feel indistinguishable from collaboration with someone who is not.
Even when it is not the same kind of view.
We may need to learn how to tell the difference between a mirror that challenges us and a window that contradicts us.
Alison + Wiggins