Uncertainty Is More Expensive Than Change

In nearly every AI adoption workshop I run, someone offers the same explanation for why their organization is struggling: people resist change. It’s a convenient diagnosis, one that puts the challenge on employees, as if adoption would take care of itself if people were just a little more open-minded.

Decades of behavioral research say otherwise, and so does my own experience sitting across the table from hundreds of employees. Most people don’t resist change. They resist uncertainty.

That’s a subtle distinction, but it changes how you approach an AI rollout. People change jobs, move across the country, get married, and pick up new skills constantly. We’re remarkably capable of adapting. What we struggle with is making decisions when the consequences are unclear, and few workplace technologies generate more unclear consequences, faster, than AI.

Our brains are wired to avoid ambiguity

Behavioral economists call this ambiguity aversion. Faced with a known option and an unknown one, most people instinctively favor the known, even when the unknown option might serve them better. We overestimate the risks of uncertain outcomes and undervalue their potential benefits. It’s an old survival instinct: unfamiliar environments once carried hidden danger, and caution kept us alive.

Our workplaces look nothing like the ones that instinct evolved for, but the instinct hasn’t gone anywhere. Every new system introduces questions that take mental effort to resolve, and AI introduces more of them, faster, than almost any tool I’ve watched organizations adopt.

AI creates questions before it creates value

When a company rolls out Copilot, ChatGPT Enterprise, Gemini, or Claude, leadership tends to lead with capability: summarize this meeting, draft that document, analyze this data. Employees care about those capabilities eventually. But before anyone evaluates the benefit, they’re quietly working through a different set of questions. Can I trust this answer? Am I allowed to use this for this particular task? Could I be exposing something I shouldn’t? Will my manager assume I’m cutting corners? If the AI gets it wrong, who owns that mistake?

Those questions almost never show up in a training deck, and yet they’re often the reason someone never opens the tool at all. The technology isn’t creating the resistance. The unanswered questions are.

Training solves the wrong problem

The default response to slow adoption is more training, and I understand the instinct. But most technical training answers “how do I write a good prompt” or “where do I find this feature,” and those aren’t the questions holding people back. The real questions are about judgment and norms: should I trust this output, when is it appropriate to use AI here, what should never touch an AI tool, will I get in trouble if I get it wrong. Until someone answers those, the uncertainty doesn’t move, no matter how many workshops you schedule.

Governance is a confidence tool, not a compliance one

This is where I’ve become increasingly convinced that governance, done well, is one of the strongest levers you have for adoption. Governance has a branding problem. People hear the word and think restrictions and approval chains. Done right, it does the opposite: it reduces the number of decisions an individual has to make alone. A clear acceptable use policy tells employees what they can safely do. Data classification guidance tells them what belongs in a conversation with an AI tool and what doesn’t. Documented use cases show people where AI already creates value, instead of asking every employee to discover that for themselves. None of that is bureaucracy. It’s friction removal.

People learn from watching leaders, not reading policy

Written policy only gets you so far, because people don’t primarily learn from documents. They learn by watching each other, and especially by watching leadership. When senior leaders use AI openly in meetings, talk about what worked and what didn’t, and model reasonable judgment instead of performing perfection, uncertainty starts to shrink. Employees aren’t just learning how to use the tool. They’re learning what’s considered normal.

The opposite is just as visible. When leaders announce a big AI initiative and then never touch the tool themselves, employees fill that gap with assumptions, usually less generous ones than reality: maybe leadership doesn’t really trust it, maybe it’s only for certain teams, maybe they’re watching how we use it. Left unanswered, people write their own policy, and it’s almost always more restrictive than the one you intended.

Small wins beat big promises

Most AI rollouts open with ambitious language about transforming the business. That messaging might be accurate, but it also raises the uncertainty it’s trying to overcome, since now everyone is wondering how their job is supposed to fundamentally change. I’ve had far more success starting small: summarize a meeting, tighten an email, organize research notes, draft a first pass at a proposal. Each small win answers a question no slide deck could: maybe this actually helps me. Confidence builds through experience, not presentation, and every successful interaction quietly replaces one unknown with one known.

Adoption doesn’t improve because people become more familiar with the technology. It improves because every clear policy, every modeled behavior, and every small success removes one more piece of ambiguity from the picture. Get enough of that uncertainty out of the way, and you stop needing to convince anyone to change. They’ll have what they actually needed all along: enough confidence to take the first step.