Why Formal Training Misses So Much Real Capability Building

Recently, I ran a vibecoding workshop where two participants completed it at roughly the same level. They built similar web applications, gave similar feedback, and their knowledge quiz scores looked about the same. From the point of view of the classroom, the learning outcome looked equal.

Two weeks later, it clearly was not. One participant had gone home, kept tinkering, built two more small web apps, hit bugs, asked AI stupid questions, learned new terms, and kept going. The other got stuck at the first serious error and stopped. If I only measured the workshop itself, I would have said they grew by the same amount. If I measured what happened after the workshop, I could not honestly say that anymore.

That is the problem with how many education systems and talent development systems still think about learning. We measure the clean bit in the room. We ignore the messy bit that happens after, even when the messy bit is where capability actually forms.

That messy bit is what I mean by invisible learning. It is real learning, often capability-shaping, but it rarely gets captured because it happens between formal checkpoints: in friction, in experimentation, in side conversations, in failed attempts, and in the quiet adjustments people make while trying to do real work.

For a long time, that learning was hard to preserve without adding too much admin. That is why I think this moment matters. The end of invisible learning does not mean total surveillance or life-logging. It means we can finally capture more of the moments that matter before they disappear, and use them to deepen judgment, support learners, and recognise capability more honestly.

The classroom is often the trigger, not the whole event

I spend a lot of time in classrooms, and I still believe in them. This is not an anti-classroom argument. The classroom is good at a few things that matter. It creates shared language. It speeds up knowledge transfer. It gives people a structured starting point. In fast-moving fields, that still matters.

But the classroom was never the whole story. It is often the trigger event, the moment that gets someone started, gives them enough confidence to try, or helps them see what is possible. What happens next is where things get interesting.

The World Economic Forum's Future of Jobs Report 2025 says 63% of employers see skills gaps as the main barrier to business transformation. Its January 2025 press release on the same report adds that, out of every 100 workers globally, 59 are expected to need training by 2030. That tells me something simple. Learning can no longer be treated as an occasional event.

And that is exactly what the workshop example above reveals. Inside the session, both participants looked successful. They could follow the content, complete the exercise, and leave with similar-looking outcomes. But that was only the classroom slice of the story. After the workshop, one participant went back, kept tinkering, and built two more small web applications. The other got stuck and gave up.

So was the workshop successful? In one sense, yes. It helped both participants get started. But in the fuller sense, it only translated into continued application for one of them. One kept learning after the session. The other stopped learning when friction appeared.

That is why I think we often define the learning event too narrowly. The workshop is not the whole event. It is the trigger. The whole event includes what happens after, when someone tries to apply what they learned in the flow of work, gets stuck, improvises, persists, adapts, or walks away. That is where capability either deepens or stalls, and that part is often the least tracked.

If learning is now constant, then a model that recognises only scheduled instruction is too narrow for the world we are actually in.

I felt that personally when I built my own computer from scratch. I did not take a course. I did not have prior hardware expertise. I had a goal, some budget, YouTube, search, AI, and a growing list of questions I did not know enough to ask at the start. Over two weeks, I learned what motherboards do, how cooling choices affect performance, how to compare GPUs, what RAM actually is, and which mistakes could get expensive fast. That learning was real, but it did not happen because someone designed a curriculum for me. It happened because I cared enough to keep going.

In many domains, the session starts the learning, but lived application is what turns exposure into capability, and that too is learning, often uncaptured.

What our current systems still mismeasure

Many formal systems reward what is easiest to standardise. They favour legibility over fidelity. If the learning happened in a lecture, a workshop, or a structured assignment, we know what to do with it. We can timebox it, assess it, document it, and file it somewhere. But if the learning happened through a failed attempt, a side conversation, a messy prototype, or repeated problem-solving over two weeks, many systems struggle to recognise it.

That is not because the learning is less valuable. It is because the learning is harder to tidy up afterwards. The OECD's current work on informal learning makes this explicit. Informal learning is increasingly essential, yet still under-recognised and undervalued in policy and measurement frameworks. The OECD's adult learning overview says the same thing more bluntly: formal and non-formal training are only the tip of the iceberg, and a lot of learning is acquired through work experience and informal exchanges at work. In other words, a large share of learning is still effectively invisible to the systems that claim to track development.

That is exactly the mismatch I keep seeing. We are currently working with an institute of higher learning to rethink how learning outside the classroom can be captured more meaningfully. As part of that work, we have been building Brainframe, an AI-supported reflective learning companion designed to help learners capture, structure, and revisit meaningful learning moments. One issue kept surfacing: reflection written too late becomes story-editing. By the time learners are asked to explain what they learned, they often reconstruct a cleaner version of events than what actually happened. The polished report may satisfy the institution. It does not necessarily preserve the learning trail.

One university leader said something close to this in a conversation that stuck with me:

Students can always get a polished output from a generic AI tool.
What disappears is the chain of learning that produced it.

That line matters because it shifts the question. The real issue is not whether someone can generate a respectable final artefact. The issue is whether the learner, the assessor, and the institution can still see how judgment formed.

When we only look at neat reflections written at the end, we miss how the learning actually happened.

The real skill is the ability to build new skills

This is the part I think many organisations still underinvest in. We say we want capable people. Then we treat capability as if it is mostly about content transfer. Often it is more about whether someone knows how to keep learning once the class is over.

Go back to those two vibecoding participants. The workshop gave both of them a similar starting point. What separated them later was not just knowledge. It was the willingness and ability to keep building after the formal learning event ended. That is why I keep coming back to a simple idea: one of the most important skills in the AI era is the skill of building new skills.

That skill usually depends on three things:

• A reason to learn.
• A way to learn.
• A felt need to keep going when it gets frustrating.

When I built my computer, all three were present. I wanted the machine. I had tools and resources. I also had enough skin in the game to push through confusion. Without that mix, I probably would have stopped.

This matters for talent development because many organisations still jump too quickly to content delivery. They buy the workshop before they build the conditions for learning. Then they wonder why attendance happened but behaviour did not shift. The pressure is this: the World Economic Forum says 63% of employers already see skills gaps as the main barrier to transformation, and 59 out of every 100 workers are expected to need training by 2030. The harder question is whether our systems are helping people become more self-directed and more resourceful, or whether we are still overvaluing the moment of instruction because it is the easiest part to count.

In a world where tools and explanations are abundant, the durable advantage is not access to content but the capacity to keep learning from use.

Not everything should be captured

Once people hear this argument, some immediately jump to total capture. That would be a mistake. The answer is not to log every conversation, every thought, every failed attempt, and every passing reaction. That creates noise, raises privacy concerns, adds cost, and turns reflection into admin.

There is a practical question here too: how do you turn real learning inside a person into something a system or AI can actually work with? That takes resources. It takes time, prompts, workflows, and some form of support. If you try to capture everything, you do not just create bad data. You also create unnecessary cost and learner fatigue.

So the real question is much narrower: what is worth capturing because it would otherwise disappear, and because keeping it would improve future judgment or performance? Not every moment should be recorded. Some moments are private. Some are trivial. But some are exactly where growth happens, and learners need to get better at noticing them. For me, the best candidates are usually:

• moments tied to a meaningful problem
• mistakes that changed how someone approaches the next attempt
• small pieces of advice that reshape behaviour
• patterns that only become visible across multiple experiences
• decisions where the reasoning matters as much as the outcome

A mentor once told me to stop saying words like just and quickly in presentations because they weaken the point before it lands. Tiny comment. Lasting effect. That kind of learning usually vanishes because it does not look dramatic enough to document. But those are often the moments that change how a person shows up at work.

This matters at the organisational level too. People are developing real capabilities outside formal programmes all the time, and much of it never gets seen. No one knew I could build a computer. No one certified me in AI. But I studied it, applied it, and now it is central to my work. The learning happened. The capability is real. The problem is that the system never captured it.

So no, I do not want total life-logging. I want more disciplined capture of the moments that actually compound.

The future of learning capture is not exhaustive recording. It is selective preservation of moments that improve future action.

Tacit knowledge matters even more when explicit knowledge gets cheaper

For years, explicit knowledge dominated because it was easier to teach, easier to assess, and easier to scale. If you can write it down, you can distribute it. That is still useful. But AI changes the economics of explicit knowledge. Instructions, summaries, first drafts, checklists, and explanations are now much easier to generate and access than before.

When explicit knowledge becomes cheaper, the premium shifts. It shifts toward judgment. Taste. Interpretation. Timing. Pattern recognition. Knowing what matters in this situation, not just in theory. In other words, it shifts toward tacit knowledge.

The OECD's 2019 paper on returns to different forms of job-related training found that informal learning is by far the most common form of job-related learning at work. It also found that informal learning is associated with 3.5% higher wages, and that workers in more autonomous, team-based environments are 12% more likely to experience informal learning. That is worth sitting with. The most common learning at work is often the least formal, and the environments that produce it best tend to give people room to act, interact, and figure things out.

I saw another side of this recently in a conversation with someone in PR. We were discussing which parts of her work might be turned into an agentic workflow. When I asked how she decides which trends and articles matter, her answer was basically, "I just know when I see it." That is a real answer. It is also a tacit one. The real work, then, is not merely capturing explicit steps. It is helping people surface the judgment inside their steps.

As AI floods the world with explicit guidance, the scarce and valuable layer becomes the tacit reasoning people struggle to articulate.

What AI makes newly practical

This is where I think AI becomes genuinely useful for learning. Not because it can do the learning for us. Because it can reduce the friction around noticing, questioning, structuring, and remembering.

McKinsey's January 2025 report, Superagency in the Workplace, found that nearly half of employees want more formal AI training and see it as the most helpful intervention for increasing day-to-day AI use. That is important, but I do not read it as a simple call for more classroom hours. I read it as a sign that people want support close to real work, where adoption rises or falls.

That is the practical opening. AI can help capture source material while someone stays present in the conversation. It can prompt reflection while the memory is fresh. It can ask follow-up questions that draw out tacit reasoning. It can connect one learning moment to previous ones. It can preserve the context around a polished output so the output is not the only thing left. This is what makes the end of invisible learning newly practical. The issue is not that we suddenly discovered learning outside the classroom. The issue is that we now have better ways to notice it before it vanishes.

That is why I do not think the goal is just a smarter notebook. A useful system should help deepen learning while it is being captured. It should help a learner notice where they are, what changed, what they are missing, and what this moment connects to. That is very different from using AI to generate prettier reports after the fact.

AI is most valuable when it stays close to the learning moment and strengthens reflection, not when it only beautifies the final write-up.

What a better learning system should actually do

If we accept that more meaningful learning happens outside formal instruction, then a better system should be designed around that reality.

It should be low-friction, because heavy process kills honest reflection.

It should be close to the moment, because delay turns evidence into reconstruction.

It should be selective, because not every experience deserves storage.

It should create linked memory, because isolated notes rarely show growth patterns over time.

It should support learner ownership, because systems built only for compliance die the moment the programme ends.

And it should still produce workplace-useful outputs, because managers and talent teams do need practical ways to see progress, spot capability, and support development. Ignoring that operational reality would be naive.

That combination is why Brainframe became interesting to us. Not because the world needs another journaling tool. Because the real gap is guided reflection with context memory, shaped in a way that still produces something usable for the people evaluating learning. The product is not the main argument. The pedagogical shift is.

Better learning systems should help people turn lived experience into reusable judgment without forcing them to become full-time archivists of their own lives.

The shift we need to make

I do not think the future of learning is classroom versus non-classroom. That framing is too shallow. Learning has always gone beyond the session. Students do homework. People learn while trying to get real work done. The classroom is only one part of a much longer learning process.

That matters because many workplace programmes are still designed as if the workshop is the main event. It is not. The real test comes after, when someone has to work alone, apply the idea, struggle through the messy parts, make mistakes, and still move forward. Those are not side effects of learning. In many cases, they are the real learning opportunity.

If that is true, then instructional design has to go further than the one- or two-day session. Organisations need better ways to create those follow-through moments, make them safe to reflect on, and capture the capability being built without turning managers into full-time observers. Managers already have work to do. The system should help learners surface what they tried, where they got stuck, what they learned, and what they need to explore next.

That is where Brainframe becomes interesting to me. At its best, it does not just store reflection after the fact. It helps deepen the learning while it is being captured, and it can point a learner toward the next relevant question, gap, or area to explore. In that sense, it starts to act less like a notebook and more like a learning GPS.

That is the shift I care about.