AI stories are generally written into one of two camps. The first is transformation: productivity doubled, costs halved. The second is cautionary: the chatbot hallucinated, the implementation cost three times the budget and nobody is quite sure what it delivered. Both are plausible. Less often acknowledged: both can be true of the same organisation at the same time.
What gets left out of almost every headline is the part in the middle: the slow, unglamorous work of getting people to genuinely change how they work. That part rarely makes the news. It is also the part that determines whether any of this sticks.
ADKAR is one of the most widely used frameworks for individual change. It breaks the process into five stages: Awareness (understanding why the change is needed), Desire (actually wanting to participate), Knowledge (knowing how), Ability (being able to do it in practice), and Reinforcement (keeping the new behaviour in place). It predates AI by decades. It maps almost perfectly onto why AI rollouts fail.
The easy trap is to jump straight to Knowledge. Book a training session, introduce the tool, expect people to start using it. But if Awareness and Desire are not already in place, Knowledge does not land. People sit through the training and go back to doing things the way they always did, not because they are resistant to change, but because the case for it was not compelling enough.
The Desire stage, in ADKAR’s framework, tends to come from seeing someone you trust do something with AI that you recognise as genuinely useful. A peer solving a real problem from your own kind of work, not a vendor demonstration from another industry. That is why internal advocates matter in AI rollouts: not as cheerleaders, but as credible proof that the effort is worth it.
The organisational layer
Kotter’s change model works at a different level. Where ADKAR describes what individuals need, Kotter describes what organisations need: build urgency, form a guiding coalition, develop a clear vision, communicate it, remove barriers, generate visible early wins, consolidate progress, and anchor the change in how the organisation actually works.
AI initiatives that struggle often do the first step well, building a sense of urgency by pointing at competitors or productivity data, and then jump straight to implementation. The steps in between are where change actually happens, or quietly fails to. The early wins step, in particular, is underrated. People need to see something work before they are willing to believe it will work for them. That win does not have to be dramatic. It just has to be real, visible, and close enough to home to be meaningful.
The dogs that don’t bark
Jeff Bezos borrowed the image from Sherlock Holmes: in a mystery story, the dog that did not bark in the night was itself the clue. The dangerous signal is not always the one you are hearing. Sometimes it is the one you are not. In an AI rollout, a quiet team is not necessarily a settled one. More often, silence means people have quietly reverted to doing things the old way, or are struggling with something they do not feel comfortable raising. It looks like compliance. It rarely is. The only way to know is to ask, directly and regularly.
This connects to a tension Amazon built into its own leadership principles: Think Big (do not limit your ambition to what is comfortable) against Bias for Action (stop planning and start building, because most decisions are reversible). Both instincts are right. Think Big without Bias for Action produces a strategy that never starts. Bias for Action without Think Big produces a collection of disconnected experiments that generate activity without changing anything that matters. The organisations that handle this well hold both at once.
The quieter failure
There is also the opposite problem, which Smart Company has written about: what happens when AI is adopted too readily, without intention. When teams reach for AI by default rather than by deliberate choice, something starts to erode. The judgment that came from doing the work. The skills that people stop practising because the tool handles it now. It shows up gradually, in output that is competent but thin. That is not a reason to avoid AI. It is a reason to be clear about when and why you are using it.
What changes things
The organisations that handle this well tend to share a few things. They start with a specific problem, not a technology decision. They build Desire before they build Knowledge. They generate early wins that are visible inside the team. And they treat AI as part of how people work, not a replacement for the judgment that makes the work worth doing.
That is the thinking behind the sprint.
References: ADKAR framework (Prosci / Jeff Hiatt and Timothy Creasey); Kotter’s 8-step change model (John Kotter); Smart Company, “Neural Notes: the hidden cost of using too much AI in business.” MIT Professional Education, Applied Agentic AI for Organizational Transformation (2025).