6-minute read.
It can help, but only if it’s the rehearsal and not the stand-in. The best way to encourage healthy disagreement at work is to make disagreeing part of the job, let people practise it out loud where the stakes are low, and react well when someone actually does it. AI is genuinely useful for the first two. It can build the strongest case against your plan in seconds and give you something to argue with. What it can’t do is stand up in the meeting and say “I think we’re wrong” for you. That part your team has to learn on its feet.
What happens when people let AI do the disagreeing?
Harvard Business Review’s piece on AI and psychological safety, by Jayshree Seth and Amy Edmondson, opens with a scene most managers will recognise with a wince. A leader spends six months building the case for a regional expansion and presents it to his team. Two days later an anonymous email arrives: a three-page document in which someone has asked an AI tool for the strongest case for the plan, and the strongest case against it.
The objections existed. Somebody in that team had doubts strong enough to go away and build a document. What they didn’t do was put their name to it.
We don’t think the AI is the story there. The silence is. The tool did what it was asked, and probably did it well. But the team now has a critique with no author, nobody to question, and no one who has practised standing behind a difficult point.
Why is anonymous dissent a warning sign?
Because it tells you disagreeing out loud feels too expensive. And it’s a very common feeling.
63% of US workers say they use AI to avoid difficult conversations at work, and 44% say they sometimes freeze in face-to-face conversations because they can’t review or edit their words first.
(Preply survey of 1,002 employed US adults, carried out with Dataframe, reported June 2026)
That second number is the one that worries us. If the polished, edited version of you only exists on a screen, the live version gets less and less practice. Then the meeting comes, someone pushes back, and you’ve got nothing to say.
In our sessions we often ask people to argue the side they don’t agree with. A participant at an accountancy firm told us afterwards that “arguing for points I don’t believe in made me slow down and think more about my points than when it is something I already agree with.”
That’s the muscle an anonymous email never builds: hearing the other side out loud, then answering it out loud.
What is AI actually good for here?
Quite a lot, if you’re honest about the job you’re giving it.
Use it to find the hole in your plan before your colleagues do. Ask it to play the sceptical finance director, or the client who thinks you’re too expensive, and argue back. Ask it what the strongest objection to your proposal is, then decide whether you’ve got an answer. All of that is preparation, and preparation is where most people skimp.
We built our own AI coach, Deb, for exactly this kind of practice. Deb gives feedback on four things: clarity, composure, influence and impact. It’s there at the point of need, the night before the hard conversation, when there’s no facilitator in the room. But it’s human-led, AI-enabled, and the order of those words matters to us. The rehearsal happens on a screen. The performance doesn’t.
What do you lose if the machine argues for you?
You lose the practice. And practice is the whole thing.
Forbes ran a piece last month by career coach Kelly C. Huang that quotes a Goldman Sachs partner worrying about “cognitive atrophy”: outsource your reasoning to a model often enough and you stop doing it yourself. Her concern is junior staff, who used to learn by doing the work under supervision and now risk skipping straight past it.
Disagreement works the same way. Nobody learns to hold their nerve when a senior colleague says “I’m not sure that’s right” by reading a better script. They learn it by doing it badly, in public, until they’re good at it. A team that lets AI raise every objection is a team that never builds that skill, and on the day the stakes are real, nobody in the room has practised saying the awkward thing.
So here’s the line we’d draw. AI drafting the counter-argument is useful. AI delivering it, anonymously, instead of a person is a symptom.
How do you encourage healthy disagreement at work?
01 Make it a role, rather than a risk.
Before a decision, ask one person to make the strongest case against it, and rotate the job. The person pushing back is now doing what they were asked rather than being difficult, which is exactly how a debate works.
02 Let people switch sides.
Ask the person most in favour of a plan to argue against it for two minutes. It feels like a game. It isn’t. A graduate at a global energy company told us that being asked “to present a case for the opposing viewpoint” made them realise “how useful it is to practise more active empathy.”
03 Use AI for the homework, then have a human say it.
If someone has used a tool to find the weak spots in a plan, great. Ask them to bring the two strongest points to the meeting and make them out loud.
04 Keep the stakes low at first.
Nobody’s first go at challenging a director should be in front of the board. In our sessions, people are on their feet early, in small groups, with at least two facilitators for every 15 participants. The point is to make the first attempt happen somewhere it doesn’t cost anything.
What should you watch for in your next meeting?
Your own reaction to the first person who disagrees with you properly. Thank them. Answer the actual point. If they’re right, change something, visibly. Your team will decide whether disagreement is safe based on what happened to the last person who tried.
FAQs
It’s a signal worth taking seriously. The concerns are usually real; what’s missing is the confidence that raising them openly won’t cost anything. Treat the anonymous critique as useful information, then work on making the next objection safe to say in person.
AI is good at playing the other side: the sceptical stakeholder, the unhappy client, the colleague who disagrees. Rehearsing your reply out loud against a strong counter-argument is useful preparation. It works best when the real conversation still happens face-to-face.
Give disagreement a job. Ask one person per meeting to make the case against the plan, rotate the role, and respond well when they do. People copy what gets rewarded, so the first few reactions matter most.
