After my last post about AI and the slightly identity-confused Defender running on electricity, several conversations started circling around the same question: “Okay. But what are organisations actually underestimating when implementing AI?”

Here are five challenges I believe will define successful AI adoption over the next few years.

#1 AI is not an IT project. It’s a behavioural project.

Most AI initiatives begin with platforms, governance, security and use cases. Which makes sense. However, adoption does not happen in technical planning spreadsheets. It happens in human behaviour.

AI changes how people make decisions, collaborate, evaluate quality, build confidence or stop feeling confident. And if organisations do not actively work with the layers that drive human behaviour, they rarely achieve the adoption they expect. It is a bit like January gym memberships – We may show up a few times. But are we actually becoming fitter?

#2 You cannot train people out of uncertainty.

A common organisational response to AI anxiety is: “Let’s run more training.” And yes — training matters. Training helps bridge gaps in functional knowledge and more training makes the unsafe familiar and reduces friction. But no amount of training sessions can fully remove deeper uncertainty: Will I still be relevant? What will I be measured on now? What is still “my” work? When am I good enough? What happens if I cannot keep up?

People can fully understand AI and still feel deeply threatened by it. That does not make them resistant. It makes them human.

#3 Middle managers may become the most overlooked risk group.

Executives are excited. Specialists are experimenting. Consultants are making slides with purple gradients. But middle managers? They are standing directly between efficiency pressure and human stability.

They carry many of the same fears and uncertainties as their teams. But in addition, they also face growing uncertainty around their own expertise, authority, coordination and decision-making role — while simultaneously being expected to create calmness, clarity and motivation for everyone else. That is not one additional layer of complexity. It is several. And it is a fairly brutal combination.

#4 AI enthusiasm is not the same as adoption

A large number of employees will likely “perform AI” long before they truly adopt it. They will say: “Yes, we use it”, “Super exciting”, and “Absolutely, makes total sense” – while quietly continuing to work much as before.

Not necessarily because they are resistant or incapable. But because most people quickly learn what the organisation wants to hear.

Performative adoption — talking about AI integration while primarily maintaining existing behaviours — may become one of the biggest blind spots in AI transformation. A bit like using MS Teams for meetings while still running the organisation mainly through email.

Real behavioural change requires energy, cognitive capacity, psychological safety, time, repetition and new habits. And honestly, I still rarely see any programme investing as heavily in the human infrastructure as required to support that kind of change

#5 Silence may be more dangerous than resistance

Traditionally, change programmes watch closely for signs of resistance. But in AI transformations, silence may actually be far more alarming. Because if people are still trying to understand what AI truly means for their role, relevance and future, is it really “safe” to say that uncertainty out loud?

Silence is easily mistaken for alignment. But when it comes to AI it is often uncertainty, observation, quiet hesitation, or social self-protection that is hidden in the unspoken.

Though AI may be changing the tools, humans still operate on the same psychological mechanics we always have. And for roughly 300.000 years, caution toward the unknown have played a central role in human survival. The challenge is simply that technology develops far faster than psychology does.