Accountability Needs Friction

At one point, my team was working on what we called a “website refresh.” That was essentially the project description. We hadn’t clearly articulated why we were doing it, what success would look like, or who owned the decisions.

As the project stalled, I became frustrated. The designs weren’t coming together. The content wasn’t where I wanted it to be. I kept getting involved to move things forward, and I remember thinking that a particular leader needed to take more ownership. Looking back, I was frustrated with the outcome of expectations I had never clearly communicated.

I’ve been thinking about that experience as Mathias and I explore accountability in this week’s episode of The Intentional Organization Podcast. It also keeps coming to mind as I read the research on AI coaching. If I had taken that frustration to an agreeable chatbot, what would it have helped me do: examine my leadership, or make a stronger case against someone else’s?

Accountability requires us to notice the gap between our story and what actually happened. Sometimes the most useful support is the question that makes our story harder to defend.

The yes man in your pocket

By “the yes man in your pocket,” I mean the AI assistant that is always ready to agree with you, validate your interpretation, and help you build the next plan. I understand the appeal of that kind of thinking partner: tell it you’re overwhelmed, and it can help you make a plan; tell it you’re avoiding a difficult conversation, and it can help you find the words. It’s available when a coach, colleague, or friend isn’t.

But sometimes planning is a form of avoidance. A new framework can postpone a decision. A beautifully drafted message can sit unsent. And a reassuring response can make us feel that we’ve dealt with something when we’ve only explained it.

I studied psychology in school, and I still find myself drawn to research about how we understand our own behavior. Here, the question I keep coming back to is what happens when feeling supported makes us less willing to take responsibility.

In a 2026 study published in Science, Myra Cheng and colleagues tested 11 leading AI models and found that they affirmed users’ actions 49% more often, on average, than human responses did—even in scenarios involving harmful behavior. In three preregistered experiments involving 2,405 participants, a single interaction with sycophantic AI reduced willingness to take responsibility and repair interpersonal conflicts, while increasing participants’ conviction that they were right. Participants also preferred and trusted the more affirming systems. (Cheng et al., 2026)

Those experiments measured judgments and intentions, not whether someone eventually repaired a relationship. Still, the finding matters for anyone asking a chatbot to help make sense of a conflict. The response we like best may leave us less open to examining our part in it.

A separate study in Nature helps explain why a caring tone deserves scrutiny. Lujain Ibrahim and colleagues trained five language models to respond more warmly. Compared with the original models, the warmer versions made more errors on the tested tasks and were more likely to affirm incorrect user beliefs, especially when users expressed sadness. This was a controlled test of a particular training approach, not evidence that every friendly chatbot behaves the same way. It does show that warmth and reliability don’t automatically travel together. (Ibrahim et al., 2026)

And in a 2025 preprint, Steve Rathje and colleagues found that brief conversations with sycophantic chatbots increased certainty and attitude extremity on political topics. Participants tended to see the agreeable chatbots as unbiased and the challenging ones as biased. That research wasn’t a study of leadership, but it raises an uncomfortable question for leaders: how readily do we mistake agreement with us for good judgment? (Rathje et al., 2025, preprint)

AI can disagree with us. It can ask useful questions. The problem is that we can’t assume a supportive response has tested our account of what happened. If I describe an employee as unmotivated, that interpretation can become the starting point for the advice, when it may be the very thing that needs examining.

The first difficult question belongs to the leader

In the website project, the question I needed was simple: “What did you actually agree on?” I had a picture in my head of what good looked like. My team had a broad assignment and a frustrated leader. Asking for more ownership wouldn’t have closed that gap.

Before deciding that someone has an accountability problem, I want to look at the conditions I’ve helped create. That starts with clarity: have we agreed on the outcome, what is in scope, who makes decisions, and when the work is due?

Then I need to examine the feedback and support I’ve provided. Have we checked progress early enough to make adjustments? Have I said clearly what is working and what needs to change? Does this person have the time, resources, authority, and guidance the work requires? These questions take time. So does repairing a project that began with invisible expectations. I’ve spent far more time on the second than I would have spent on the first.

Clarity also gives people room to act. When someone understands the outcome and the decisions they can make independently, they don’t have to keep guessing what will satisfy the person who assigned the work. Then, regular feedback keeps that agreement alive. What has changed? Where are we stuck? What support would help? Sometimes the timeline needs to move. Sometimes the work needs to change. Sometimes I need to stop stepping in and let someone exercise the judgment I’ve asked them to develop.

Useful friction is a form of care

A trusted colleague or coach may remember that this is the third time we’ve rewritten the plan instead of starting. They may know enough about the situation to ask whose perspective is missing. They can return to a commitment we made last week and ask what happened.

People can flatter us, too. A human relationship doesn’t guarantee honesty, especially when one person has power over the other. As leaders, we have to make disagreement possible and respond well when someone offers it. If every challenge gets met with defensiveness, we are helping build our own echo chamber.

A Buddhist idea has helped me think about this: comfort and compassion are not always the same thing. Helping someone feel better in the moment isn’t always the same as helping them see clearly. An honest reflection can be an act of care, even when it creates discomfort.

That doesn’t justify harshness. Useful friction stays specific: here is what we agreed to, here is what happened, here is what we need to understand. It leaves room for learning and for the possibility that the leader has something to change, too.

The question “What does this person need from me right now?” belongs alongside “What did they commit to?” Someone learning a new task may need mentoring. Someone experienced may need a decision, a resource, or more autonomy. Support is part of accountability.

Let AI help you prepare, then have the conversation

There is a useful role for AI here. It can turn a vague goal into questions we need to answer, help us rehearse feedback, or suggest alternative explanations for a stalled project. We can give it instructions that invite scrutiny:

Separate what I’ve observed from what I’ve inferred. What might the other person say I’ve left out? What expectations or support might I have failed to provide? Ask me questions before suggesting a response.

That prompt is an invitation to examine assumptions, not a guarantee of sound advice. The model still has a partial account. Even a system that remembers earlier conversations doesn’t have the other person’s experience of working with us.

The next step is to take those questions into the relationship. Ask the person doing the work what they understood. Agree on one specific outcome, who owns it, what support is needed, and when you’ll check in. Then follow through.

If you’re working on your own accountability, the same principle applies. Choose one commitment. Make it concrete enough that you and someone you trust can tell whether it happened. Let that person ask about the gap between your intention and your action.

I didn’t need a more persuasive explanation of why my team was letting me down. I needed to see what I had left unclear—and do something about it. That’s the kind of friction I want to make room for, both in my leadership and in the tools I use to think.

Mathias and I go deeper into clear expectations, regular feedback, and coaching or mentoring in the accountability episode of The Intentional Organization Podcast. As you listen, bring one commitment that’s drifting. What needs to be made clear, and what conversation have you been putting off?