The AI adoption events I attended were, in most respects, well-organised and genuinely interesting. Good presentations, compelling use cases, enthusiastic energy about what is possible. I left most of them with the same nagging feeling: the hardest part of AI adoption was barely being discussed.

What follows is not scepticism about AI's potential. It is a set of observations about the challenges that tend to determine whether AI initiatives actually produce value — and that were mostly absent from the conversations I was part of.

Putting AI everywhere is a mistake

One of the first risks in AI adoption is the tendency to treat it as a general-purpose solution. Not every problem benefits from automation. Not every process improves by adding an intelligent model. Without clear criteria for where AI makes sense and where it does not, organisations end up investing in applications that add noise rather than signal.

The useful question is not what AI can do — it can do a great deal — but which specific business problem you are actually trying to solve, and whether AI is the most direct route to solving it.

AI is not neutral — it redistributes power

AI changes who has access to relevant information, who can interpret it, and who makes decisions based on it. This creates internal political dynamics that organisations rarely address explicitly. Many forms of resistance to AI are not technological in nature — they are resistance to changes in control, status, and internal narratives about who knows what.

Organisations that treat AI adoption as a technical deployment and are surprised by the political friction are missing the most predictable part of the challenge.

The adoption is inherently asymmetric

AI adoption does not happen at the same pace across an organisation. Some people quickly amplify their capacity with new tools. Others fall behind. This creates internal gaps that are not always visible but are highly consequential. People who were previously similarly capable are suddenly operating at very different levels of output.

If this asymmetry is not actively managed, AI creates new forms of organisational inequality that compound over time. The individuals falling behind are often not the least capable — they are the ones who were least supported in the transition.

Resistance tends to be quiet

Active resistance to AI is rare. Passive resistance is common: tools that are technically available but practically unused, decisions that are delayed without explanation, initiatives that are superficially supported but never quite get the conditions they need to succeed. Behind much of this is fear — of losing relevance, of professional identity being undermined, of becoming the person who did not keep up.

This kind of fear does not respond well to training programmes that focus on tool capability. It responds to honest communication about what is actually changing, genuine support for the people most affected, and visible evidence that the organisation values the human contribution that AI cannot replace.

What organisations should prioritise

A few things consistently matter more than others in AI adoption that actually produces value.

Start with more structured work, where cause and effect are clear, before applying AI to complex work where the dynamics are harder to understand. This builds confidence and competence in conditions where mistakes are less costly.

Define what success looks like before you start, not after. The organisations that struggle most with AI adoption are often the ones that launched without a shared understanding of what they were trying to achieve or how they would know if they had achieved it.

Create genuine spaces for experimentation — places where people can try, fail, and learn without consequences. Adoption that only happens in controlled demonstrations stays in demonstrations.

The final observation is one I have become increasingly confident about: AI adoption is one more test of organisational culture. Organisations with genuinely agile cultures — not the ones that use the vocabulary but the ones that have actually built the habits of learning, adaptation, and honest conversation about what is working — are consistently better equipped to navigate this transition than those without. The tools are not what determines the outcome.