This is a perspective shaped by direct experience rather than theory. Having worked with organisations at various stages of AI adoption, the pattern I keep seeing is this: the technical challenges get solved. The human challenges do not.

Two strategies, two different failure modes

Most AI implementations I encounter fall into one of two broad strategies.

The first is automation to reduce headcount: machines replacing human work, often framed diplomatically but understood clearly by everyone in the organisation. The message lands before any communication is sent. The predictable responses follow — resistance, passive non-cooperation, a thousand small acts of friction that collectively undermine the efficiency the strategy was meant to achieve.

The second strategy — producing more with the same number of people — is generally more sustainable. But it comes with its own set of anxieties. People facing this scenario are not afraid of redundancy. They are uncertain about what their role will become, what new expectations they will face, and whether they have the skills to meet them. If those questions are not addressed explicitly, resistance emerges in quieter, less visible ways. It is harder to diagnose and harder to address.

The common denominator: uncertainty

What both strategies share is uncertainty. AI introduces ways of working that are less deterministic than what came before. Outcomes cannot always be defined in advance. The normal frameworks for planning and accountability start to feel unstable.

How an organisation responds to that uncertainty determines whether AI adoption creates real value or stalls. In my experience, this is where the decision is actually made — not in the selection of tools or the design of workflows.

What reduces uncertainty

There are a few practical decisions that consistently improve the conditions for sustainable AI adoption.

Start with clearer, more structured work. Introducing AI into highly complex work first — where cause and effect are hard to trace and outcomes are inherently ambiguous — increases friction and confusion. Starting with work that is more deterministic makes it easier for people to understand what the AI is doing, why it is doing it, and how it fits into their day-to-day work. This builds confidence before the harder applications are attempted.

Define success explicitly before you start. It is surprisingly common for AI initiatives to launch without a shared agreement on what success actually looks like. When the criteria are not explicit, interpretation is open — and ambiguity generates anxiety. Being specific about what problem is being solved, what indicators will measure impact, and how you will distinguish real improvement from side effects creates a foundation for honest assessment.

Support the people whose work is changing. Every automation initiative changes someone's day-to-day experience. People need to understand clearly what is changing in their role, what will be expected of them, and what support is available. This is not about removing all discomfort — change is uncomfortable, and pretending otherwise is not helpful. It is about creating conditions where people can act with clarity rather than confusion.

Treat adoption as iteration, not installation. AI adoption is not a one-time transformation. The organisations that do it well run experiments, observe results, adjust, and iterate. The ones that treat it as a project to be completed tend to find that the system they installed does not actually match the organisation's reality.

The incentive problem

Even when all of the above is done thoughtfully, outcomes can still fall short if incentives are misaligned. Under pressure, people act in line with what is measured and rewarded — not what is stated in a change programme. The unwritten rules of the system often pull harder than the explicit ones.

This is why AI adoption, at its core, is a cultural challenge as much as a technical one. It closely resembles what many organisations discovered during agile transformations: new practices were introduced, but existing incentives kept pulling behaviour in the old direction. The technology changes faster than the culture. Closing that gap is where the real work happens.