All articles
Unlearning Is the Skill of the Decade. Nobody Budgets for It.
September 15, 2026 
Every organisation that books me to speak wants its people to learn AI faster than the competition. Almost none of them has a plan for what those people need to forget.
The training budgets are real. Prompt courses, AI academies, champions in every department. Yet when people find me in the hallway after a keynote, their story has little to do with missing skills. They know how to open the tool. What holds them back is twenty years of doing it the other way, inside a culture that still rewards the other way.
Your brain solved this problem a long time ago
In 2017 the neuroscientists Blake Richards and Paul Frankland made a striking case: forgetting is part of how memory is designed to work. The brain actively weakens and overwrites old memories, because the purpose of memory was never to archive the past. Its purpose is to make better decisions in a world that keeps moving. A brain that kept everything would cling to details that no longer apply. It would stay loyal to a world that has already gone.
Organisations have the remembering half and almost none of the forgetting half. Every approval step, reporting line and KPI is the memory of a problem someone once solved, and nobody prunes them. Bo Hedberg wrote in 1981 that knowledge grows and goes stale at the same time, so understanding means taking in the new and throwing out what has become misleading. Forty-five years later we still budget for the first half only.
Why unlearning is harder than learning
Old knowledge does not wait quietly in a drawer while the new knowledge moves in. It blocks the door. In a well-known series of experiments, chess players faced positions with a familiar solution and a better, less obvious one. The familiar move pulled even masters away from the best answer, dropping them to the level of players three standard deviations below them in skill. Their experience showed them the old move first.
Habits run deeper still. Wendy Wood’s research found that around 43 percent of what we do each day is habit, performed in the same setting while the mind is elsewhere. A 2025 study in Psychology & Health put the share of behaviour triggered by habit closer to two thirds. A training programme talks to the conscious part of the day, and the habits run the rest.
BCG’s 10-20-70 rule says the same: roughly 10 percent of AI success comes from algorithms, 20 percent from technology and data, and 70 percent from people and processes.
The part leaders miss
Change needs safety before it needs skills.
Edgar Schein, who spent his career at MIT studying culture, described two anxieties inside every change. Survival anxiety says you must change or fall behind. Learning anxiety is the fear of what changing will cost: looking incompetent for a while, being punished for it, losing status, losing your sense of who you are. People move only when the first outweighs the second, and Schein’s advice was to shrink learning anxiety instead of inflating survival anxiety.
Most AI kick-offs do the reverse. The slide says adapt or die, and the room hears die.
Fear then produces the very behaviour leaders want gone. Organisational researchers call it threat rigidity: under threat, people and systems narrow what they take in, pull control to the top and fall back on their best-rehearsed routines. Threaten people and the old habits come back stronger. We ask them to change while making change dangerous, and then call the result resistance.
You can see it in the data. A December 2025 report by Infosys and MIT Technology Review Insights found that 83 percent of business leaders say psychological safety directly affects the success of AI initiatives. Only 39 percent rated their own as high.
On the other side of the desk, people are hiding. When Microsoft and LinkedIn surveyed 31,000 people in 2024, more than half of those using AI at work were reluctant to admit using it for their most important tasks, and 53 percent worried it made them look replaceable.
Listen closely to that word, replaceable. The fear underneath it is older than any job description. It is the fear of losing meaning.
In research by BetterUp, published in Harvard Business Review, more than nine out of ten workers said they would trade part of their pay for more meaningful work, on average 23 percent of their future lifetime earnings. So when a machine writes in thirty seconds the report someone took pride in for fifteen years, the question in their head goes far beyond job security: if a machine can do this, what was I for?
Researchers call it professional identity threat, and a 2024 study in AI & Society found it lowers people’s willingness to adopt AI, with seasoned experts among the most exposed.
This is where technology starts to make us more human. It lifts away the tasks we had confused with our identity and confronts us with a question most careers never had to face, about where our value lies once the task is gone. Leaders who help people answer it get change. Leaders who skip it get polite compliance on the surface and quiet sabotage underneath.
The interface has stopped being the bottleneck
For decades we learned the language of machines. Command lines, menus, thick manuals. Jakob Nielsen called AI the first new way of working with computers in more than sixty years: you describe the outcome you want and the machine works out the steps. ChatGPT passed 900 million weekly users earlier this year, and hardly any of them opened a manual. Technology is climbing down towards the oldest interface our species owns, which is conversation.
When the tool speaks human, the remaining friction is human too. Microsoft’s research found people needed about eleven weeks of steady use before AI became part of how they worked. Nobody needs eleven weeks to learn how to type a sentence. Eleven weeks is how long it takes to replace a habit.
McKinsey found employees are three times more likely to be using generative AI than their executives believe, and concluded that the biggest barrier to scaling it sits with leaders. Your people are further along than you think. What many of them are waiting for is permission to let the old way go.
Where unlearning starts
So start with the unlearning, and make it safe. Every time you add something new, say out loud what people may stop doing. Go first yourself, in public, by naming a belief you have dropped and what replaced it.
Protect the stretch of temporary incompetence that every real change requires, because that is when people decide whether trying was worth the risk. And anchor meaning in the contribution people make instead of the tasks they perform, so that losing a task never feels like losing a self.
A caterpillar does not become a butterfly by adding wings to its old body. Inside the chrysalis it breaks most of that body down, keeping small clusters of cells that carry the plan for what comes next. It is the most radical act of unlearning in nature, and it happens inside the safest shelter the animal will ever build. Your people need that shelter before they need another course.
Over to you.
What did you have to unlearn to get where you are?
I’m also curious where you land on the bigger argument. In your organisation, is the real bottleneck unlearning and safety, or is it still skills and tools? Tell me in the comments. I read every one.
Christian Kromme writes The Human Spark – Beyond AI, a newsletter about what it means to be human in a world being reshaped by technology.