The Gap Between What We Expect From AI and What People Experience

Most conversations about AI in the enterprise are conversations about capability. What can the model do. What can we automate. What can we deploy next quarter. These are not unreasonable questions, but they are the wrong place to look if you want to know whether an AI initiative will actually work. The more useful question is quieter and harder: once the system is in place, what actually changes for the people doing the work, and does it match what anyone expected?

That gap, between what organizations expect AI to do and what their people actually experience, is the center of my doctoral research. It is also, in my experience, the center of why enterprise AI so often disappoints despite working exactly as designed.

Two things that are not the same

An organization adopts AI with a picture in mind. Faster decisions, better targeting, less manual work, sharper performance. That picture is the expectation, and it is usually built from the technology's capabilities, what the system can do in principle.

The people on the ground encounter something else. A tool that changes how their goals are set, how their work is measured, what judgment they're still trusted to exercise, and what now happens automatically whether they agree with it or not. That is the experience, and it is built not from capability but from consequence, from what the system does to the texture of the actual job.

When expectation and experience line up, adoption feels natural and the initiative compounds. When they diverge, you get the familiar pattern: a technically successful rollout that somehow fails to produce the change it promised, and a workforce that has quietly routed around it. The technology didn't fail. The gap did.

Why goal-setting is where this gets sharp

One place this tension becomes especially concrete is goal setting, which is where my first line of inquiry sits. For decades, our best understanding of how goals drive performance has rested on a fairly stable picture: specific, challenging goals, freely accepted and supported by feedback, tend to produce better performance than vague encouragement. It's one of the most durable findings in organizational psychology.

Intelligent systems disturb that picture in interesting ways. When an AI assists or even sets the goal, recalibrates it in real time, and tracks progress continuously, several of the quiet assumptions underneath the old model are suddenly in question. Is a goal still freely accepted if a system assigned it? Does feedback still motivate the same way when it's constant and automated rather than periodic and human? Does a moving target, adjusted by a model faster than a person can internalize it, still function as a goal at all, or as something else wearing a goal's clothes?

I'm not going to claim the answers here; that's what the research is for, and the work is ongoing. The point I want to make is narrower and I think more useful: a forty-year-old theory of motivation was built for a world where humans set the goals, and we are now changing who, or what, sets them. That is exactly the kind of place where expectation and experience come apart, because the capability ("the system can set and track goals") and the consequence ("what that does to how a person experiences their own work") are genuinely different things.

Why this matters to people who aren't writing dissertations

The reason I study this practically rather than abstractly is that the gap is expensive, and it's avoidable. Leaders adopting AI tend to underwrite the expectation and underestimate the experience, and then they're puzzled when a capable system doesn't land. The fix isn't more capability. It's paying attention to the human consequence early, designing for it, and being honest that a tool which changes how work feels is a tool that will be adopted or rejected on those terms, not on its technical merits.

The aim, in the end, is practical: helping leaders adopt AI without quietly dismantling the things that made their organizations function in the first place. You cannot do that by studying what AI can do. You can only do it by studying what happens to people when it does.