In 2016 I wrote an article criticizing how organizations approach learning and development. The target was the factory model: management decides what skills people need (as if such a thing were knowable), experts assemble a workshop and a slide deck, employees are extracted from their work, processed through the training, assessed, and returned to their desks officially “developed.” My argument was that this model belonged to a mechanistic world that no longer existed, and that learning needed to be integrated with real work, chosen rather than mandated, and organized around real problems.
I stand by all of it. But I’ve been overtaken by events, and so has the factory model — not reformed, but obsoleted by something much stranger than anything I predicted. The interesting question in 2016 was how should organizations train people? The interesting question now is: when machines can do the work people used to learn by doing, where will the next generation of judgment come from?
The part of L&D that just became worthless
Start with what AI has already done to the old paradigm, because it’s more total than most L&D functions have admitted.
The factory model’s core product was knowledge transfer: moving articulable information and procedures from someone who has them to someone who doesn’t. Workshops, courses, certifications, the LMS — an entire industry built on the premise that transferring knowledge is difficult and therefore valuable.
That premise is now false. Anything that can be articulated — concepts, procedures, frameworks, examples, practice problems, feedback on the practice problems — can be delivered instantly, personalized, conversationally, on demand, at essentially zero marginal cost. The patient tutor who never judges you, never tires, and meets you exactly at your level is available to every employee at every moment. Whatever you think about AI, this much is settled: the transfer problem is solved. A corporate function organized around solving it is a candle factory after electrification.
This sounds like the end of an argument. It’s actually the beginning of one, because it forces a question the training paradigm let us avoid: if knowledge transfer was the easy part all along, what was the hard part?
What training never produced
Think about the most capable people you’ve worked with — the ones you’d want in the room when something genuinely hard or genuinely weird is happening. What makes them valuable is almost never their stock of articulable knowledge. It’s something harder to name: they know which of the seventeen plausible framings of a problem is the real one. They sense which risks are theatrical and which are live. They can feel, before they can prove, that a plan smells wrong. Call it judgment, discernment, taste — the developed capacity to perceive and decide well under conditions where no procedure applies.
Here’s the inconvenient truth the training industry spent decades stepping around: judgment cannot be transferred. It is grown, and it grows in only one soil — cycles of doing real things, with real stakes, getting real feedback, and reflecting on what happened. The seasoned engineer’s intuition is compressed experience: thousands of decisions, hundreds of mistakes, each one metabolized into perception. It’s the difference between someone who has read about music and a conductor who hears, in the same symphony, things the rest of us cannot detect. Nobody lectured the conductor into that. They accumulated it, rep by rep.
This was the real argument buried in my 2016 piece — that learning had to be integrated with actual work because the classroom never produced the capability that mattered. I framed it then as an efficiency claim. I’d now frame it as an ontological one: the work is the curriculum. There is no other curriculum. There never was.
The apprenticeship problem
Which brings us to the genuinely new thing — the development that makes this an urgent work design problem rather than an L&D modernization project.
If judgment grows only through doing, then every profession’s judgment pipeline has always rested on an unglamorous foundation: routine work. The junior lawyer develops legal judgment by grinding through document review. The junior analyst develops financial intuition by building models cell by cell. The junior engineer develops taste by writing the boring code and watching it fail. The junior consultant develops client sense by drafting the deck a partner will tear apart. Routine work was never just output. It was the apprenticeship — the thousands of low-stakes reps through which novices slowly became the people you want in the room.
AI is consuming exactly that work, first and fastest. The tasks most easily delegated to machines — the structured, the repetitive, the well-specified — are precisely the tasks that constituted the bottom rungs of every judgment ladder. And here’s the part that should keep work designers up at night: delegating those tasks is individually rational every single time. The AI does it faster and cheaper, this quarter’s numbers improve, and the cost — a junior who never develops, a judgment pipeline quietly running dry — lands five to ten years in the future, on someone else’s watch. It’s a tragedy of the time horizon. Each delegation decision is sensible. The sum is an organization that has liquidated its capacity to grow senior people, discovering the problem only when the current seniors leave and there is no one behind them who ever did the reps.
I want to be careful here, because the convenient conclusions are all wrong. The answer is not refusing the technology — that’s competitive suicide. Nor is it pretending the old apprenticeship survives intact — it doesn’t. The honest position is uncomfortable: organizations now have to do deliberately what the economy used to do for free. Development used to be a byproduct of production. It is becoming a separate thing that must be designed, protected, and paid for on purpose — or it will simply stop happening.
Designing for judgment on purpose
So what does that look like? The prescriptions from my 2016 piece survive, but each one comes back transformed by the new situation.
Integrate learning with work becomes something stronger: reserve developmental work for humans, deliberately, even when a machine could do it. This will feel like inefficiency, because locally it is — the same way a flight simulator is an inefficient way to get from Newark to Denver. The point of the simulator was never transportation. Organizations will need to identify which categories of work are their judgment gymnasiums and staff them with developing humans (working alongside AI, with seniors reviewing) as a conscious investment, not an oversight. The org that automates 100% of its junior work has not become efficient. It has eaten its seed corn.
Organize learning around real problems becomes: make the experiment your unit of development. Every genuine organizational experiment — a removal trial, a new way of working, a strategic probe — is a complete judgment cycle in miniature: frame a hypothesis, act with real stakes, confront real feedback, decide what it means. An organization running a steady portfolio of deliberate experiments is operating a judgment academy without ever calling it one. This was always true; it’s now the most reliable apprenticeship structure left standing.
Treat people as autonomous adults survives intact, with one addition. Mandatory training was always the fastest way to make people hate learning, and a buffet of self-selected development beats a prescribed menu — that hasn’t changed. What’s new is that the autonomy must extend to the human–AI boundary itself: people need real agency in deciding which parts of their work to delegate and which to keep because keeping them is how they grow. An employee with no say in that boundary has no say in their own development.
And one prescription I underweighted in 2016 now carries the whole structure: build the reflection infrastructure. Experience does not automatically become judgment — it becomes judgment only when attention is paid to it. The reps count only if they’re metabolized: retrospectives that actually happen, debriefs with people further down the road, an operating rhythm with noticing built into it rather than bolted on. In an environment where raw experience is getting scarcer, the conversion rate from experience to learning becomes the number that matters most. Reflection is how you raise it, and reflection is made of nothing but protected attention.
The organization that learns is the organization that lasts
I’ll end with the strategic claim, because this is no longer a soft topic.
When every organization has access to the same models trained on the same accumulated knowledge, articulable capability is flat — table stakes, identical across competitors. What cannot be downloaded is precisely what this whole piece has been about: the judgment of your people, individually and — just as important — collectively, in the form of an organization that knows how to notice, make sense, and decide together. That capability is grown, slowly, through reps and reflection, inside your specific context, and it compounds: the more developed your people’s perception, the more they notice, and the more they notice, the faster they develop. It is the one asset on the field that a competitor cannot acquire by purchasing the same subscription you did.
In 2016 I closed by saying the goal was organizations that have learned how to learn. I’d sharpen it now: in a world where knowing has been commoditized, learning is the last durable advantage — and it will belong to the organizations deliberate enough to keep growing judgment after the economy stopped doing it for them.
A note on lineage: an earlier version of this argument appeared in 2016 as “The Future of Getting Better at Work," written during my decade at The Ready. The critique of the training paradigm and the case for learning embedded in real work began there; the commoditization argument, the apprenticeship problem, and the design prescriptions for the AI era are where the thinking has gone since.