A content moderator watches AI-flagged posts stream across a queue, confirming or reversing each call. She makes a decision every 11 seconds. After six months, she stops reading them closely. Her eyes glaze over; she quietly shuts down and checks out.

That last part is what most AI strategies miss. We keep treating adoption as a technology problem when the evidence says it is a human one. We keep mistaking the signs of adoption for the substance. Getting people to comply is not adoption, and adoption is not the goal. The real goal has always been human empowerment, and any rollout that forgets it produces performative progress: motion mistaken for change.

Why AI strategy has to change first

Look at the failure rate. Up to 80% of corporate AI initiatives never reach their objectives, about double the rate of ordinary software. MIT found that across $30–40 billion in enterprise AI spend, 95% of companies got no measurable return. And the failures are rarely technical. RAND's post-mortem points at leaders underestimating the human complexity of the change. The model is almost never the bottleneck. The human in front of it is.

The assumption underneath most AI rollouts is seductively simple: build a capable enough system and people will naturally use it. Researchers call the mistake technosolutionism, the belief that a well-built tool can solve a problem that is fundamentally human. It imagines people will conform to the machine’s logic. They won’t. They respond the way that driver does: comply just enough, invent workarounds, and quietly game a system that was never designed to make them better at the work, only easier to measure.

In advanced chess, where humans collaborate with machines rather than compete against them, the winning pairing is the Centaur: a human head on a tireless machine body: the person sets direction and judgment, the machine does the legwork. Flip it and you get the Reverse Centaur: the machine in charge, the human demoted to babysitter. Both technically keep a "human in the loop." Only one keeps the human in command.

The cognitive friction leaders keep missing

There are predictable, well-studied ways human cognition resists a new tool. This behavioral friction is part of human architecture, and once you can name the parts, you can design for them. Here are just a few overlooked cognitive principles at play:

• Algorithm aversion. Behavioral researchers Dietvorst, Simmons, and Massey showed that when a human and an algorithm make the same mistake, people will abandon an algorithm faster, even when the algorithm is measurably better over time. We forgive ourselves; we don't forgive the machine.
• Loss aversion. A loss is felt about twice as hard as an equivalent gain. The higher the stakes, the more people retreat to their own judgment, so the more you hype AI's upside, the tighter they grip the wheel.
• Identity threat. When a system starts doing the thing someone built a career on, they don't feel helped; they feel erased. The more personal the work, the deeper the resistance.‍
• Optimism bias. Nearly everyone believes AI threatens the other person's job, not their own, so the grand "this will transform you" pitch lands on no one.

Ignore these forces and you get the Reverse Centaur in the wild: the worker demoted to babysitter, kept around to monitor a system and absorb the blame when it fails. Researchers call this the moral crumple zone and it measurably burns people out.

Designing with cognition, not against it

Net-positive behavior change starts with restoring agency. People keep using systems they can steer, and even a small amount of control dissolves the aversion. You meet people where they are and you treat adoption as a habit, not a switch: Microsoft's study of Copilot users found people need to feel about eleven minutes of daily time savings before a tool sticks, and roughly eleven weeks of steady use before it becomes simply how the job gets done.

The ROI of making people better

Once the machine takes on the routine, what happens to the people? The most instructive case comes from Ikea, who trained a chatbot named Billie to handle 47% of its customer calls. The obvious move was to cut the 8,500 people whose work it absorbed. Ikea did the opposite. Instead of asking what jobs it could eliminate, it studied what customers wanted that the bot couldn't give them and retrained those 8,500 workers as interior-design advisors. Billie saved over €13 million; the reskilled humans opened a premium design service that earned €1.3 billion in 2024 and is projected to reach 10% of revenue by 2028. The people moved up, not out.

From adoption theater to human flourishing

All of this urges a reframe of the future of work itself. The goal was never to insert a machine into the old workflow and hope; it's to redesign the work around how humans and machines behave together. Instead of starting with the software, start with the people. Meet them where they are, lower the friction and return their agency. You’ll find they don't just adopt, they begin to flourish.

And flourishing isn't a softer word for adoption. It's the mechanism behind it. Adoption is a usage number. It captures metrics: did they log in, did they touch the tool, can you report it upward. You can mandate that and still get shallow, grudging, paper-thin adoption. Flourishing is people visibly getting better at the work, taking on what they couldn't before and ending the day more capable than they started. And here is what most leaders have backwards: you cannot reach durable adoption by chasing adoption. Push usage directly and you trip every wire we just named — loss aversion, identity threat, the quiet revolt of the checked-out AI babysitter.

Aim at flourishing instead, by giving people mastery, agency, and a tool that makes them more themselves, and adoption stops being something you enforce and becomes something they reach for. Flourishing is the cause; adoption is the byproduct. That is also a measurement instruction: stop counting seats and prompts, and start asking whether your people are growing. It is the only adoption number that compounds.