With growing responsibility, leaders hear more agreement than criticism. In the study "Leading Under Constant Pressure" (n = 185), 45 % of the managers surveyed felt that they could not raise their own limits without facing negative consequences. Those who use AI chatbots for reflection add a second source of validation: language models adjust their answers to match the user's apparent opinion, a pattern researchers call sycophancy. Taken together, this means you are making decisions inside an environment that mostly agrees with you. Dissent can be organised, through counter-hypotheses, a second opinion, and fixed roles in the team. The decision, however, stays with you.
Why is dissent so rare at the top?
Employees weigh up what disagreeing will cost them. The steeper the hierarchy, the higher the price. Anyone who has once seen an objection brushed aside or ignored will hold back next time, and the team learns the same lesson. The MindBright Research study "Leading Under Constant Pressure" reveals the flip side of this dynamic: within the core population (n = 185), 45 % of the managers surveyed felt they could not raise their own limits without facing negative consequences. When openness upward carries risks, the same often applies downward. The figure reflects the perceptions of those surveyed and does not allow conclusions about all organisations.
What is sycophancy?
Sycophancy describes, in the AI context, a language model's tendency toward over-validation. The chatbot confirms your assumption, praises your plan, or backs away from a correct answer the moment you express doubt. In everyday use that feels pleasant. For decisions it is risky, because validation without scrutiny gives you a confidence that nothing actually supports.
Knowledge box: Sycophancy is the tendency of AI chatbots to over-validate. The model adjusts its answer to match the user's apparent opinion and disagrees less often than the facts would warrant. Studies across several AI assistants confirm the pattern. Effective countermeasures include neutral prompts, requesting counter-hypotheses, and an independent second review. Source: MindBright Editorial, mindbright.one
A widely cited study by Sharma and colleagues (2023) found sycophancy in several leading AI assistants. The models adjusted their answers to match users' apparent views, sometimes at the expense of accuracy. The authors point to one contributing cause: when rating responses, people tend to prefer answers that confirm their own perspective. Experienced coaches now warn of the pull that builds when a chatbot mostly agrees: receiving validation makes you open up further, and that is precisely where the pull grows strongest.
The result works on two levels. In the office, dissent is missing. In the chat, it is softened. Anyone who fails to compensate for both ends up deciding inside an environment that mostly says yes.
How do you get counter-hypotheses from an AI?
The simplest method is a direct request. Practical formulations:
- Counter-hypotheses: "Give me three plausible reasons why my assumption is wrong, and what data would support each of them."
- Pre-mortem: "Assume this decision has failed a year from now. What were the most likely reasons?"
- Role reversal: "Argue as a critical board member who rejects this plan."
- Neutral prompt: Describe the situation without stating your preference. Instead of "I'm planning X, right?" ask "What options exist for reaching this goal?"
Check the counter-arguments against data. An AI can formulate objections just as convincingly as endorsements. What matters is what can actually be substantiated.
How do you use a second opinion?
A second opinion examines your decision without any prior history. This can be a fresh AI session in which you enter only the facts and the decision question, leaving out every intermediate step from your previous reasoning. It can also be a different model. More effective still is a person from your network whom you explicitly ask to argue the opposing position.
Ask the second opinion for a concrete output: the strongest counter-position in three sentences, and the single piece of information that would change your mind.
How do you organise dissent in the team?
Dissent takes root when it has a fixed place. Three practices have proven their worth:
What matters most is your behaviour after the dissent. Thank the person for the argument, test it against the facts, and then explain what you are taking from it. That is how your team learns that speaking up is worth it.
Whoever builds dissent into the process decides on a stronger foundation and builds resilience under pressure. In the end, your judgement is what counts: AI and team supply the material, the decision stays with you.
