AI-aware and capability-poor: the most expensive mistake we are making in L&D

Most of the strategic conversation about AI in organizations is about work. What will be automated, what will be augmented, what new roles will emerge, how productivity will shift. These are pertinent questions and I have no quarrel with them being asked.

But underneath them, something quieter is happening that I think matters more for the next five years. AI is changing the infrastructure of learning itself — where capability is built, what kind of capability we now need, and how that capability actually forms in people.

Most organizations have not yet registered this second shift — and the few that have being few kind of proves my point. Right now, most organizations I encounter and observe are responding to it with the only instrument they know: more programs about AI. More literacy. More fluency. More prompt training. I think that response is solving for one variable in a three-variable problem, and it’s going to leave a lot of well-intentioned organizations with workforces that are AI-aware and capability-poor at exactly the moment they can least afford it.

Let me lay out the three shifts, and then what I think the actual response is.

What we learn is shifting — about, with, through.

There are three distinct modes of engagement with AI when it comes to learning, and most organizations are operating in only one of them.

The first is learning about AI — literacy, awareness, vocabulary. Necessary, pretty easy to scale, and where most of the budget currently sits. It produces people who can talk about AI more or less competently. It does not necessarily produce people whose work has changed because of it.

The second is learning with AI — and this is where the conversation gets confused, because the phrase is being used to mean three quite different things.

In its weakest sense, “learning with AI” means using AI. People access the tool, get outputs, complete tasks faster. That is genuinely useful, but it is not learning in any meaningful sense. The person doing it is not developing. They are operating — ideally, they are gradually operating more efficiently.

In a stronger sense, it means a process in which the person’s own thinking is being sharpened through their work with the tool. They are learning to brief, to specify, to interrogate an answer rather than accept it. The dialogue with the tool is making the quality of their reasoning visible to them in a way nothing else has, and they are responding to what they see. Their questions have become better. Their tolerance for vague work — including their own — has dropped. (Case in point: I know colleagues who have been working with AI daily for a year and who write, think and brief more sharply than they did before. I also know colleagues who have used it just as much and write worse than they did before — because they stopped doing the messy thinking that used to happen between draft one and draft two. The tool was the same. The learning was not.)

And in a third sense, increasingly important and still poorly understood, “learning with AI” means engaging AI as a deliberate development partner. Not the tool you use to draft an email, but the coach you have structured conversations with, the tutor that meets you where you are in a domain, the reflection companion that nudges and challenges and remembers. The new generation of AI coaching, tutoring and companion applications is a fundamentally different category of product, even where it’s running on the same underlying model. The intent is different, the design is different, and the kind of development that becomes possible is different.

Most organizations have collapsed these three meanings into one. They have invested heavily in giving people access to general-purpose AI and lightly in the conditions that turn access into learning — reflection, productive friction, visible practice in teams — and barely at all in the deliberate use of AI as a development partner. Without those distinctions, “learning with AI” becomes a polite phrase for “using AI.”

That conflation is the single most expensive mistake in most of corporate L&D right now.

The third mode is learning through AI — using the encounter with AI as a forcing function to develop genuinely human capabilities the tool itself cannot provide. Judgment under ambiguity. Taste. The ability to ask the second question. The ability to lead a team whose ways of working are quietly being rewritten. This is the layer almost no one is talking about, and it’s the one I think matters most.

A clarification before we move on. There is, of course, a fourth thing AI is doing in this space, and it is real: AI is reshaping how learning itself is designed, delivered and personalized. Adaptive curricula, intelligent tutoring, real-time facilitation support, learning platforms that meet people where they are, experiential designs that would have been logistically impossible a few years ago. This is significant work, and the organizations doing it well will pull ahead. I am not arguing against it. I am arguing that even the most sophisticated AI-powered learning systems do not, on their own, produce the deeper human capabilities the moment demands. They are part of the answer. They are not the whole answer. The rest of this piece is about what they do not solve — which is most of what matters.

Where we learn is breaking down — the 70/20/10 is under pressure on all three fronts.

The model that has anchored corporate learning for three decades — 70 percent on the job, 20 percent from others, 10 percent formal — is fracturing in all three places at once.

The 70 is the most worrying. The messy entry-level work where people used to accumulate judgment is being automated. Junior people now arrive at problems that used to take five years to encounter. They have more guidance available than any generation before them. They also have fewer chances to struggle productively, which is where capability actually forms. (Case in point: that’s precisely why millennials don’t just fix the printers of their parents but also their kids’. Because they never had to learn that type of technical problem-solving.)

The 20 has changed shape. “The others” now includes artificial beings. Powerful ones. But a copilot is not a colleague, and the social fabric of learning — apprenticeship, side-of-desk mentoring, the corridor conversation — is thinning in ways most organizations haven’t reckoned with. (Case in point: in most organizations, budgets for the above — apprenticeships, physical gatherings, offices and schedules encouraging idle time — are on the decline.)

The 10 is being asked to do something it was not traditionally designed for. Formal learning used to be about transferring knowledge, maybe developing skills. Now it has to cultivate posture — how someone holds themselves in the face of what they don’t know. Most curricula haven’t caught up or are stuck in limbo.

Case in point: the type of ROI question we get exposed to on the regular speaks to the degree of delusion some people have that learning is still about transferring knowledge or building skill as if we are ticking boxes on a training calendar. What you know or can do is of decreasing relevance; it’s what you are able to sit with, reason about and sustain, despite pressure that is more and more the question.

How we learn needs a wider lens — and this is where the response actually sits.

If the what and the where are both shifting, the instinctive response — more AI programs, better AI-powered platforms — addresses neither in full. AI programs are a partial answer to what. Smarter learning systems are a partial answer to how. Neither is an answer to where, and neither is sufficient on its own.

The actual response is to lean hard into three modes of learning that have been quietly underfunded in corporate L&D for two decades, and that AI — however sophisticated — does not replace.

Contextual learning. Most corporate development is inward and present-tense. The richest capability development happens when people are pulled out of their own context and forced to read signals from elsewhere — other industries, other functions, other geographies, other moments in time. Looking inward, outward, backward and forward at once. This is how people develop the ability to recognize patterns they have never personally lived through, which is exactly the capability AI cannot give them and the situation they will increasingly need to lead in. AI is extraordinary at synthesizing what is already known. It is not a substitute for the felt sense of having stood somewhere genuinely unfamiliar.

Experiential learning. Not learning about something, but learning through it. There is a difference, and senior audiences feel it immediately. You cannot lecture someone into a capability. You cannot prompt someone into one, either. You have to put them in a situation where the capability is the only way through. This is the mode that builds capacities — the things people can hold rather than the things they know. Tension. Polarity. Contradiction. Ambiguity over long time horizons. These don’t develop through curriculum, and they don’t develop through AI dialogue. They develop through being placed in conditions that demand them and then having the support to reflect on what happened.

Communal learning. Learning is a social act. Especially now. The lone learner with their AI companion is a real archetype, and it is not enough on its own. Capabilities form in friction between people. They consolidate in conversation. They get tested in groups before they get trusted in roles. The organizations whose people are developing fastest right now are not the ones with the best individual tools. They are the ones whose teams have learned to learn out loud together — and who have protected, deliberately, the human contact that AI-mediated learning quietly erodes.

This is not nostalgia for the way learning used to be. The world the older modes were built for is not coming back, and these models need upgrading and reinvention, too. But the human capabilities they developed are now in shorter supply than ever, and the easy answers — more content, more platforms, more AI — will not produce them.

The questions I would ask

If I were a CHRO right now, I wouldn’t be asking how AI-literate the workforce is. I wouldn’t even be asking how much people are using these tools, or how AI-enabled our learning platforms are. I would be asking a different set of questions.

Where, in our organization, are people still being placed in genuinely unfamiliar contexts that stretch their judgment? Where are they being put through experiences that build capacities, not just transfer knowledge? Where are they learning together in ways that make their thinking visible to each other?

And — are we doing more of this, or less, than we were five years ago?

The honest answer, in most organizations I work with, is less. Quietly, gradually, for understandable reasons — but less. At exactly the moment the case for more has never been stronger.

That is the work. Not making people more human. Building the conditions under which human capability can keep developing in an environment where the old conditions are quietly dissolving.

Different question. Different answer.