Stop Using Carrots and Sticks to Motivate People to Use AI
Let's chat about the real problem we're trying to solve, shall we?
Last week, I was chatting with a senior leadership team before their offsite. They were planning next year’s AI strategy, and twenty minutes in, the conversation landed where it usually does. AI adoption was uneven. Their teams had the tools. AI power users were running ahead… and then there was everyone else. Leadership was growing frustrated.
For the past year, they had been working with two levers. The first lever was the carrot. More access, more training, internal demos, recognition for the early adopters. These efforts had given them power users, but left everyone else in the dust. The second lever was the stick. Adoption metrics in performance reviews, manager accountability, a clear ‘use AI or it shows up in your rating’ signal sometime in Q1.
Like many leadership teams I sit with, they didn’t see a way forward beyond this approach. They were tired of the gap between what they knew was possible and what their people were actually doing. This exhaustion is real across every industry right now. It’s the lived experience of a year and a half of brute-force AI adoption, and it’s wearing good leaders out.
Is this really an AI motivation problem?
The brute-force frame goes like this. AI is the future. Competitors are moving. We’ve bought the tools, we’ve offered the training, we’ve put it in the strategy. If our people aren’t using AI, the problem must be our people, which means their motivation, their willingness, their resistance to change. So we reach for the levers that work on people, which are the carrot and the stick.
This is the same diagnosis Microsoft, Meta, and Nvidia are running, with bigger budgets and louder mandates. Microsoft put AI use directly into reviews. Meta is grading employees on ‘AI-driven impact’ starting this year. Nvidia’s Jensen Huang said publicly that an engineer earning $500K who isn’t spending $250K on AI tokens is a warning sign.
And then there’s Duolingo. CEO Luis von Ahn went ‘AI-first’ in April 2025 and tied AI use to performance reviews. A year later, he reversed it. Employees had started asking whether they were being asked to use AI for AI’s sake. His new framing, on a podcast a few weeks ago, was that the most important thing in performance is doing the job as well as possible. If AI helps, use it. If it doesn’t, he’s not going to force the issue. That reversal is the tell.
Duolingo had usage. What it didn’t have was value, and the gap between the two is exactly where the mandate cracked.
YOU MIGHT ALSO LIKE: Why SaaS Companies Require a Bigger Mission, Not a Smaller Headcount
What Deming Saw Coming in 1986
The deepest version of this argument is forty years old. W. Edwards Deming wrote it in Out of the Crisis in 1986. His diagnosis was that the system people work in, and the interactions inside that system, account for something like 90 to 95 percent of performance.
Re-read that: the system people work in accounts for 90-95 percent of performance.
The individual contribution sits inside that. Which means when you try to motivate your way past a system problem, you spend a lot of energy and get almost nothing back.
Deming was unsparing about performance reviews tied to numerical goals. He called the idea of a merit rating ‘alluring,’ and then named exactly how it fails: ‘pay for what you get; get what you pay for; motivate people to do their best, for their own good. The effect is exactly the opposite of what the words promise.’ That was 1986. He was talking about manufacturing quotas. He could just as easily have been describing a 2026 AI usage dashboard.
His most-quoted line is “people work in the system, and management creates the system.” Which means management owns the system. The brute-force approach to AI treats adoption as a people problem, which is the one place a leadership team has the least direct control. The system is where they have the most.
When was the last time your AI strategy meeting focused on workflow design instead of adoption rates?
So What Does Management Actually Own?
A lot, as it turns out, and almost none of it is on the carrot-stick continuum.
Management owns the four dials I write about in Hyperadaptive, which are people, purpose, process, and resources. The mistake is to turn only one. Introducing new incentives without changing the workflow. New training without changes to capacity allocation. A new AI Lead without changes to decision rights. The dials are interconnected, and turning one and asking why the system didn’t move is the modern equivalent of asking why people aren’t being motivated enough.
Deliberate, judgment-led adoption looks different. It looks like Richard Thaler and Cass Sunstein’s idea of choice architecture, applied to AI. You arrange the context so the AI-integrated path is the path of least resistance:
AI tools sit as the default view on the intranet, not buried three menus deep.
AI usage shows up first in team meetings, not as the last bullet.
Manual workflows still exist for cases that need them, and they’re no longer the default.
You aren’t telling anyone what to do. You’re designing the environment so that the right thing is also the easy thing. You are designing the infrastructure so it supports ongoing AI usage.
That’s the move, and it’s the move that takes judgment. Every workflow is different. The choice architecture for a finance team isn’t the same as the choice architecture for a customer success team. You can’t mandate your way to that level of specificity. You can only design your way to it, one workflow at a time, with the people who actually do the work in the room.
The leadership team I sat with picked up on this almost immediately. They had been spending energy debating which lever would work better. They left the conversation asking what support structures encourage AI usage and workflow reinvention at every turn?
The brute-force version of AI adoption asks people to push through a system that wasn’t designed for the work it’s now asking them to do. The deliberate version starts at the other end and asks what the system needs to look like so the pushing isn’t required in the first place. One of these makes management’s job harder in the short term, because designing systems is harder than telling people to do things. The other makes management’s job impossible in the long term, because no amount of mandate produces adoption that compounds.
People work in the system. Management owns the system. In 2026, management is must design a system to support AI adoption.
The Hyperadaptive Model is one such system.
If you’re leading an AI transformation and finding yourself stuck on the carrot-versus-stick debate, this is the kind of system-design work we do inside the Running Hyperadaptive Orgnizations class. AI Lead Accelerator. Cohorts are forming for next quarter at hyperadaptive.solutions/class.
Hyperadaptive: Rewiring the Enterprise to Become AI-Native is out now from IT Revolution Press, with more on the four dials, choice architecture, and what management actually owns. Find it at hyperadaptive.solutions/book.
This is so insightful, thank you for sharing! I also used Demmings and the architecture of choice in my own book, but in the context of helping marketers working with marketing technology!