You Can't Feel Yourself Getting Better at AI

Here is a pattern I see often.

Someone does an AI course. A good one. They finish it able to explain what a large language model is, what an agent does, why the prompt matters. Then a month goes by, and their actual working week looks much the same as it did before. The knowledge is real. It just hasn't reached anything.

This isn't a lack of interest. These are people who chose to give up a weekend to learn about AI. And it happens too regularly to be about any one person - organisations have their own version of it, which is why so many have paid for the licences and the training and still can't point to much that is being done differently.

There is a distance between knowing about AI and knowing what to do with it - how to actually use it on the work in front of you. That distance is the real problem, and adding more knowledge won't solve it.

The standard response is to do another course or offer more training. But it is worth looking closely at the kind of training that works best.

BCG's AI at Work research has found that people who received five or more hours of training, with someone coaching them in person, were far more likely to be using AI regularly than people who received less. So the ingredients for success looked like multiple hours, on work you actually do, with someone providing guidance along the way.

Which isn't how you would describe a three-hour virtual course or a set of online modules. That is practice, with company. And practice raises a question of its own, because it is something you have to keep choosing to do, long after the enthusiasm of the first week has worn off.

So what makes someone keep going?

The answer I have seen, time and time again, is confidence. Someone circles the thing for months. Eventually they have a go at something small. It works. They come back the following week and they have done four more things. This is usually not about learning a new skill - they often had the skills already. What changed was that they felt able to try. In June, Gartner reported from Australian data that the strongest driver of whether employees use AI isn't training at all. It's confidence - which is exactly what I've noticed determines who keeps going once the AI Leadership Circle ends.

So confidence does what the training can't. At least for a while.

The difficulty is that confidence comes, and it goes. It is a judgement you make about yourself - am I capable of this - and a judgement like that can change as your environment does.

It happened to me in February. I had spent two and a half years working almost entirely in ChatGPT, but as Claude's capabilities accelerated I moved my whole practice across over three weeks. At first I felt genuinely on top of it. Then a new model landed. I had described AI as my intern, and later as my peer, and sitting there in late February I understood that it had become my superior. That was a confronting thing to recognise, and I wrote about it at the time.

Nothing about my skills had deteriorated in those weeks. If anything they had improved. What had happened was that the thing I was measuring myself against had moved, so my judgement reset - and my confidence went with it, because the environment I was measuring against wasn't holding still.

So if confidence can't accumulate, what does?

What I think accumulates is your understanding of what you are actually dealing with. And the clearest evidence I have for that comes from a small exercise I run in my AI Leadership Circles.

At the first session I ask people for one word describing how they feel about AI. Not what they think. What they feel. Six weeks later I ask the same question again.

The opening words almost always name a state. Curious. Trepidatious. Insecure. Conflicted. Overwhelmed. Many of the closing words don't name a state at all. Intentional. Targeted. Circumspect. Persist. I ask people how they feel, and they answer with an approach they have decided to take.

One woman arrived on insecure - an honest, and common, account of feeling behind in a world that increasingly assumes everybody is already fluent. By week six her word was excited, and her excitement had a specific cause. She had discovered that her professional expertise was what allowed her to use AI well. It was the very thing that made the output useful rather than generically plausible.

But one of my biggest lessons has come from instances where the word doesn't change at all. The word stays the same, but the meaning - and the frame sitting behind it - has moved.

One woman arrived trepidatious and left trepidatious. She explained it herself in the final session: same word, different scale. She had started out trepidatious about whether she could use the thing at all, and finished trepidatious about what it meant for the way her organisation was being led. The tunnel, as she put it, had become a landscape.

More than once I have had people arrive feeling overwhelmed and leave still describing overwhelm as their dominant emotion. They are always quick to explain that it isn't the same overwhelm. At the start it means the gap looks too large to cross and you don't yet know where to begin. By the end it means you can see how vast the landscape is. There is a real difference between the overwhelm of the lost and the overwhelm of the newly oriented. The former might struggle to take the first step; the latter is far better positioned to move forward with purpose.

In each of these examples, the emotional register was unchanged. But the territory it applied to had changed completely.

The adult education scholar Jack Mezirow gave this a name. A frame of reference, he said, is the structure sitting underneath your assumptions. It shapes what you expect, what you notice and what you feel, mostly without your knowing it is running at all. Learning either adds information inside that frame, or it changes the frame itself. A course does the first. The second happens when you build context, over time, against work that matters to you.

I have seen that “frame change” happen in real life. One of the women who had completed the Circle came to my monthly conversation group several months later. Over summer she had read a short history of AI. She hadn't done another course or picked up a new tool - it wasn't that she had built a capability. The book had given her context. She described how it had changed the way the whole thing felt: AI had been a wave crashing over her, and now it felt like a wave she was riding. She could see the shifts that had led here, going back much further than she had realised, and it felt less urgent. Less frenetic.

Nothing in her environment had changed. The models hadn't slowed down and her skills were where they had been. She was interpreting AI differently, and so it had become a different thing.

Which brings me to the part that is hardest to notice in yourself. We don't experience our assumptions as assumptions - we experience them as how things are. So when the way you are interpreting something shifts, there is no moment where you feel it happen. It quietly becomes the new normal, and you assume it was always there.

So how you feel about AI does matter, though perhaps not in the way we tend to read it. You cannot watch the frame move. But you can notice how you feel, and then again some time later, and see whether what it is attached to has changed. Taken on its own, naming the feeling gives you a starting point, and not much else. But read alongside what has been shifting around you, it can tell you a great deal more.

Which is why I would encourage you to not just notice how you are feeling, but to start paying attention to what might be driving that feeling, particularly if you are feeling uneasy.

Unease is a useful place to look. Early on it tends to be vague, and it is usually about you. Am I behind. Should I have worked this out by now. Later it becomes specific, and it is about the work. How do I know whether this output is right. What happens to my team's judgement if this becomes the default.

The intensity of the feeling may not reduce at all. I have actually seen it go up. The question is not whether the word itself has changed, but whether what it is attached to has. If the worry has moved off your own adequacy and onto the thing in front of you - then something has shifted: usually your understanding of what you are dealing with and what it now means for you.

So here are two questions to sit with.

If you had to describe how you feel about AI right now, in one word - not what you think, but what you feel - what would it be?

And then the harder one, which is the one that actually does the work: what is driving that?

Write both down and put them somewhere you will come across them again. Ask yourself the same two things in six weeks. And when you do, pay less attention to whether the word has changed than to whether the answer underneath it has.