But for me, the pearl in this particular oyster was buried on page 17, and it has nothing to do with adoption.

80% of education leaders rated their institution’s AI guidance as clear. Half of students and educators described the same guidance as either neutral or absent.

Same guidance. Opposite experiences?

My first instinct was to call it a communications failure. But the gap is too large and too consistent for that explanation to hold. I think something else is going on.

Seven in ten education leaders believe at least half their users have received AI training. Yet 77% of students and 53% of educators say they haven’t had any. That’s too large a gap to explain by communications failure alone. So what’s actually going on?

Unhelpfully, I think both the leaders and the students are right.

Here’s the category error that I think is at the heart of it: leaders are treating AI training as a thing you complete. A module. A course. Excel Macros 101, scheduled, delivered, ticked off. And by that definition, yes - the training happened.

But the people actually working with AI every day are experiencing something that has no completion state. Every time they use it, the tool has shifted slightly. Every time they think they’ve understood its limits, it surprises them - in both directions. The relationship between the person and the system is evolving in real time, in ways that no onboarding module was designed to address.

What they’re describing - whether they’d use these words or not - is less like learning a skill and more like navigating a relationship. And you don’t train for relationships. You develop them. Continuously, imperfectly, with regular recalibration as both parties change.

The relationship counselling analogy is worth taking seriously. Counselling doesn’t produce a certificate of completion. It builds a better understanding of the dynamic, the tools to navigate friction when it arises, and the ongoing capacity to adapt as things evolve. That’s a much more accurate description of what genuine AI literacy looks like than any training module I’ve seen.

Which means the confidence data in the same report suddenly makes sense in a way it otherwise wouldn’t. AI confidence is softening even as usage grows. Counterintuitive, if you think of AI as a skill - skills normally produce rising confidence as they’re practised. But if AI is a relationship, softening confidence as you go deeper is exactly what you’d expect. The more you engage with it, the more aware you become of how much the terrain is shifting. The more honest your relationship with it, the more you understand what you don’t yet know.

That’s not a training failure. That’s what a genuine, evolving relationship with a rapidly moving system actually feels like.

So the guidance gap isn’t primarily a communications problem. It isn’t even primarily a design problem. It’s a classification problem. Leaders have put AI in the same category as every other technology they’ve ever rolled out - something you learn, apply, and move on from - at precisely the moment when that category no longer fits.

The course was booked but the readiness didn’t follow. And it won’t, until organisations stop asking “have our people been trained?” and start asking something closer to “how is our relationship with AI developing?” - and build the structures to support an ongoing answer rather than a one-time one.