Many organisations measure their AI readiness in access: how many employees have a licence for the tool? In my view, that is the wrong measure. The interesting question is not who has access – but who has changed the way they work. In the writing programmes I teach, I see the same pattern again and again, regardless of industry: four phases, each with its own logic and its own leadership need.
Phase 1: Resistance is information
The first phase looks like a problem but isn't one. Some employees keep their distance from the tool; others try it once and conclude that “it doesn't sound like us”. It is tempting to dismiss the reaction as technophobia. I would suggest listening to it instead – because the resistance almost always points to something real: a professional pride worth preserving, or a legitimate doubt about quality and accountability. The most important leadership task in phase one is therefore not persuasion. It is acknowledging that the concern is about craft – and making that craft the centre of what comes next.
Phase 2: The experiment needs a fence
In the second phase, the curious start to play. Local habits emerge, private prompts, and a quiet divide between those who use the tool daily and those who never open it. The experimental phase is necessary – no one learns to write with AI from a policy – but it is also where quality swings the most. Text goes out that no one has truly read. The fluent language hides the fact that no one has taken a position. Here the organisation needs a fence, not a brake: a few clear agreements about what the machine may draft, what always requires human rewriting, and who stands behind what goes out.
Phase 3: Method makes the human an editor
The third phase is where the journey either succeeds or stalls. The experiments must be gathered into a shared method – and the core of the method is a division of roles: the model drafts; the human edits. I have explored that division in the essay on when the machine writes first, but in the classroom it comes down to three trainable disciplines: commissioning precisely (the clearer the brief, the better the draft), judging critically (fluent language is not the same as sound content) and editing with voice (the draft must sound like the sender, not like the average). Not by teaching everyone to prompt identically – but by teaching everyone to edit deliberately.
It is also in this phase that the organisation discovers how hard it is to separate writing skills from AI skills. Someone who knows what a good text must achieve – why, for instance, the short email is the hardest – quickly becomes a strong editor of the machine's drafts. Someone who doesn't becomes fast. Just not better.
Phase 4: Integration is silence
The final phase can be recognised by the fact that no one talks about the tool any more. AI is neither a project nor a mandate; it is part of how the writing gets done – with clear agreements about accountability and a shared standard for what may leave the building. The goal was never for everyone to use AI as much as possible. The goal was for the organisation to write better and faster without losing its voice along the way. A movement from tool to habit.
Questions to begin with
The phases cannot be skipped, but they can be led – and leading begins with an honest snapshot: Which phase are we actually in, and is every department in the same one? What does our resistance tell us about the craft we do not want to lose? Who edits what the machine writes in our organisation – and have they been given the tools for the role? If your organisation is working on exactly that shift, you will find structured programmes in my catalogue of AI skills courses.
Sources
The phases draw on the author's experience from writing programmes and AI training in Danish organisations.
