Some knots are load-bearing

[Part 2 in a series focused on friction and AI change – follow-up to “Why I Learned to Love the Knot“]

I know we’re supposed to be all in on removing friction from anything and everything, but friction is essential, and not just in the physical world. Comedy, love, learning, parenting, art – these things exist because of friction.

The friction is where the humanity resides.

It sounds almost like a bumper sticker. But I mean it precisely and I think the ability to understand, navigate and manage friction has never been more relevant.

Essential friction is not inefficiency. It’s signal. And it’s where human judgment (i.e., real, hard-earned, contextually sensitive judgment) gets built and exercised.

We are in a moment where AI is doing something genuinely remarkable: it’s reducing the effort (friction) required for an enormous range of tasks. Drafting, summarizing, analyzing, generating, organizing — work that used to require hours of human cognitive labor can now be compressed, accelerated, approximated. In so many ways, this is a straightforward win that we should embrace.

But there’s an important distinction that’s getting lost in the rush to adopt and optimize:

Not all friction is the same. And some of it, we cannot afford to eliminate.

There is bad friction — the kind that exists because of bureaucracy, poor process design, misaligned incentives or simple inertia. Automating past bad friction is a gift and I don’t feel guilty about letting it go.

But there is also essential friction — the kind that exists because something genuinely hard is happening. A decision with real tradeoffs. Stakeholder conversations where the surface-level ask and the underlying need are different things. Maybe it’s a moment of organizational resistance that, if you actually stop and listen to it, turns out to be flagging a real problem with your approach. A failure that, if you let it, teaches the team something no dashboard ever would.

Essential friction is not inefficiency. It’s signal. And it’s where human judgment (i.e., real, hard-earned, contextually sensitive judgment) gets built and exercised.

In AI transformation specifically, here’s what I watch get smoothed over when organizations treat all friction as a problem to solve:

The messy stakeholder conversation. AI tools can help you prepare for it, summarize it, even draft the follow-up. But the conversation itself (the one where someone finally says what they’re actually afraid of, or where two leaders realize they have genuinely different visions for the future) cannot be automated. And if you rush past it to hit your project timeline, you will pay for it later in ways that are much harder to fix.

The failed experiment. One of the most important things a team can do in an AI transformation is try something, have it not work the way they expected, and sit with that long enough to understand why. The instinct, especially with AI tools available, is to iterate so fast that no one ever really processes what happened. Speed of iteration is valuable. But learning requires some friction (i.e., enough slowness to actually absorb what the experience revealed).

…change leaders [must] distinguish between the friction that slows us down unnecessarily and the friction that is doing essential work.

The moment of resistance. Change leaders who lean away from resistance tend to find it again later, usually at a worse moment. The person who keeps raising concerns in every meeting, the team that is quietly not adopting the new workflow, the leader who signs off but clearly isn’t bought in. These are not obstacles to manage. They’re information. The friction they represent is telling you something about your approach, your communication, your design, or your assumptions. The answer is not to smooth it away. The answer is to get curious about it.

This is what I mean by the human + AI equation and why I think discernment is the variable that determines its value.

AI can do a tremendous amount. But the judgment to know which AI output to trust, which resistance to listen to, which process to redesign versus which discomfort to actually move through — that judgment doesn’t come from the tool. It comes from the person using the tool. And it comes from having developed real experience in the friction.

For change leaders, this has a direct practical implication: our job is not to eliminate all friction. It is to distinguish between the friction that slows us down unnecessarily and the friction that is doing essential work.

That requires discernment. Which requires having been in the mess before.

Which is exactly why, as AI continues to reshape how organizations operate and what change management looks like, I am paying close attention to which kinds of difficulty we are choosing to preserve and which we are optimizing away without stopping to ask what we might be losing.

The efficiency gains are real. The risk is real too. The organizations that will integrate AI most successfully won’t be the ones that moved the fastest. They’ll be the ones that knew which knots to untangle, and which ones to leave alone long enough to understand what they were holding together.