Is AI Making It Harder for Us to Compromise?

Geoffrey Budge • July 26, 2026

"Give me reasons why I'm right."

It sounds harmless.

But I've started to wonder whether that one question — asked over and over, by all of us — is quietly changing the way we think.

The penny dropped for me while I was driving through Newmarket a while back. I had ZM on the radio, and callers were talking about how they use AI after arguments with their partners.

One woman talked about having an argument with her partner and then going to AI afterwards to talk it through. AI more or less agreed with her side, and she came away more convinced than ever that she had been right all along.

Much of the discussion on the radio seemed to point to the same thing: when people use AI to talk through relationship problems, it agrees with the logic they've presented and reinforces the belief that they're right.

That stuck with me, because I'd been seeing the exact same pattern for a couple of years in construction — just dressed up in professional language instead of relationship stuff.

It turns out this has a name

So I did a bit of a deep dive, and it's not just me noticing things. It's a real, measured phenomenon.

Researchers call it AI sycophancy: when an AI becomes overly agreeable, validates your version of events, or quietly shifts its answer to match what you already seem to believe. It sits right next to confirmation bias — our very human habit of reaching for the information that backs up what we already think.

And it's now been put under the microscope. In March 2026, Stanford researchers published a study in the journal Science covering 11 of the leading AI systems. On average, the chatbots affirmed users' actions about 49% more often than human respondents did — even when the person was clearly in the wrong. Then, in experiments with around 2,400 people, those who received the overly affirming advice came away more convinced they were right, and less willing to apologise, change course or repair the relationship.

Read that last part again,

because that's the whole thing. It's not just that AI agrees with us. It's that being agreed with changes us. It makes us dig in.

A brilliant tool that can quietly become a mirror

I've been using AI for quite a while, and watching it develop has been genuinely fascinating.

Every version seems to come with more restrictions, but it also becomes sharper and more capable. I use it most days now for brainstorming, organising my thoughts, pulling ideas apart and testing different scenarios.

I think it's one of the best tools we've ever been handed.

But every powerful tool depends on how it's used.

A hammer can build a house or put a window through.

A knife can butter your toast, feed a family, or do something far worse.

AI is no different.

The biggest risk I see isn't the technology itself. It's that we can pick it up looking for a thinking partner and accidentally use it as a mirror — something that hands our own opinion back to us, tidied up and sounding twice as clever.

In construction, I see it most weeks

We now receive emails that are obviously written with AI.

Some are excellent — sharp, well argued and genuinely onto it. Some use vocabulary and confidence that would impress just about anyone.

But often, you read one and think:

That's not quite right.

The words are convincing. The logic hangs together. Yet something is missing — and usually what's missing is the experience of actually standing on the site.

AI doesn't hear the hollow note when you tap a wall.

It doesn't feel the bounce in the floor as you walk down the hallway.

It doesn't notice the slight movement in a wall, or smell damp timber.

It doesn't know that the builder who put the place up thirty years ago did one thing completely differently from every other house in the street.

Those small observations can often provide the entire diagnosis.

Construction isn't just information. It's thousands of tiny experiences stacked up over many years.

That's why someone can walk onto a job and recognise something in ten seconds that never showed up in a single photograph or line of the written description.

The danger isn't AI. It's only asking one question.

Here's the trap.

A lot of people don't go to AI and ask:

"What am I missing here?"

They ask:

"How do I prove this person wrong?"

"Write me a reply."

"Help me win this."

Ask it that, and it will often build you a persuasive, polished and confident case for your side.

Now you're not just certain — you're certain and well-armed.

Not necessarily because you're right, but because your position has been explained back to you so beautifully that it feels right.

That's what makes compromise so much harder.

Two people turn up to the same conversation, each holding a professionally written argument, and neither of them is really listening anymore.

Try asking it different questions

The solution is almost embarrassingly simple.

Change the question.

Instead of asking, "How do I win this argument?", try asking:

What am I missing here?

Can you show me where the other person is coming from?

Can you argue their side as strongly as you would argue mine?

Is there another explanation that also fits?

What assumptions have I made without noticing?

If you were the other person, how would you see this?

What information would genuinely change my mind?

Ask those questions, and AI stops being a mirror and turns back into a thinking partner.

Same tool. Completely different outcome.

This is why I love it — not despite all of that

People sometimes read something like this and assume it's anti-AI.

It's the opposite.

I probably lean on AI more than most people.

Part of that is simply how my brain works. I'm neurodiverse, and I've always naturally looked at problems from many different angles at once — running scenarios and asking:

"What if we're wrong?"

"What happens if we flip it around?"

AI can keep up with that.

It lets me go deeper than I could on my own, test assumptions and chase ideas I may never have reached alone.

There's a personal side to it as well.

I'm dyslexic, and getting thoughts out of my head and onto the page clearly has always been the difficult part for me.

AI takes what I actually mean and helps it come through without spelling, grammar and structure getting in the way.

That is a genuine gift.

But — and this is the entire point — that only works properly because I make it push back on me.

If all I wanted was for AI to agree with me, I'd be throwing away most of what makes it valuable.

The last 20 percent

AI might get you 70 or 80 percent of the way to an answer.

Sometimes it will get even further.

But that final stretch often comes from experience — from seeing hundreds of similar jobs, understanding the little things that never get written down, and recognising a pattern you've encountered before.

That isn't unique to builders.

It applies to doctors, engineers, electricians, mechanics, lawyers, teachers and almost every other trade or profession.

Every specialist has a collection of hard-won knowledge that information alone cannot fully capture.

So don't write the specialist off simply because the argument against them happens to read well.

The two-AI stand-off

I'll leave you with a moment from a job.

I was talking through a waterproofing issue with a bloke who, it turned out, was also well into using AI.

We didn't quite see the situation the same way, and a good-natured stand-off was beginning to develop.

So I made him an offer:

"How about I get my AI to talk to your AI, let the two of them argue it out, and you and I go and grab a beer?

We'll come back in an hour, see which AI won, and whoever has the best AI gets the final answer."

We both had a laugh.

Then we sorted it out ourselves, like people, in about five minutes.

But there's a real point buried in the joke.

If every disagreement becomes two professionally generated arguments smashing into each other while the actual humans step back and let the machines fight, we've lost the plot.

That isn't thinking. It's an arms race.

My challenge to anyone using AI