Most organisations are treating AI as a technology rollout. In my experience, what decides whether it works is the same thing that decides whether any change works.
Every organisation I've worked with has had a version of the same gap. What leadership says about the culture is close enough to what people experience to sound right in a board paper. It's far enough away to matter on the floor. AI hasn't created a new problem. It has found the old one faster.
Here's how it tends to look. There's a policy, written carefully and approved properly, sitting somewhere most people couldn't find if you asked them. There's a leadership message about being ambitious with AI, which lands differently depending on who delivers it and whether they use the tools themselves. And there's what people actually do. Some use it every day, quietly, because it helps and they'd rather not ask. Others won't touch it, because they're not sure what's allowed and don't want to be the one who gets it wrong. Neither group is being difficult. Both are responding sensibly to mixed signals.
I've seen this pattern before, without AI in it. When I did my MBA research at Admiral, the company values were everywhere: on the walls, on lanyards, in the annual report. Of the five people I interviewed, only one could name all four, and that was the trainer whose job was to teach them. The values weren't wrong. They just hadn't travelled. AI policies are going the same way, and the stakes are higher. The decisions AI now touches (who gets shortlisted, how performance is written up, what information leaves the building) are the ones where a gap between intent and practice does real damage.
I've also seen what closes it. When I led the people strategy and governance for Admiral's move to hybrid working across 7,500 people, the policy was never going to be what made it work. What mattered was whether line managers understood it well enough to apply it to real teams and real requests, and whether people trusted that the rules meant what they said. AI adoption is the same kind of change: a set of principles that only become real in thousands of small decisions made by managers.
That's why I start in the middle. Middle managers are the people most expected to explain AI to their teams, and the least likely to have been equipped to. If they aren't confident, they either block it or wave it through, and neither is a strategy. At Alaya, AI literacy sat inside the leadership programme rather than alongside it, because using AI well is a leadership behaviour before it's a technical skill. I use these tools in my own work every day, and that helps. It's much easier to talk honestly about where AI helps, where it doesn't, and where human judgement has to stay in the loop when you're doing it yourself.
What a second reading looks like
Before buying more licences or writing another policy, ask your own people four questions:
- Do you know what our AI policy says, and could you apply it to your work this week?
- Have you heard leadership explain what AI is for here, in a way that made sense?
- Are you using any AI tools we haven't provided? If so, why?
- What would help you use it well?
The answers will show you where the distance is, and it's rarely even. It sits with particular teams, levels or locations, and very often with the managers in between. Then close it in order. Fix the leadership signal, make the guardrails usable rather than just published, and build capability where the answers say it's missing. The technology is the easy part. It usually is.
Dan Kennedy, Second Reading