The Work Made Us

We didn’t only produce work. The work produced us. As machines take on more of the doing, the harder question is not only what we do next, but how anyone becomes capable at all.

The thing under the job

I have spent a lot of my career as the senior technologist who gets handed a junior’s work and knows, almost at a glance, that something is off. Often I can’t tell you straight away which detail is wrong. I just feel the shape of it: a missing constraint, a risk waved away a little too easily, a recommendation that looks right but doesn’t quite smell right. From the outside that instinct can look like a knack, or intuition. Really it’s just years of having done the work, settled into a feel for it.

Ask someone what they do and they’ll tell you their job. Ask who they are, and if they’re honest, a good deal of the answer is the same thing wearing different clothes. The title is the least of it. What sits underneath is the competence: what they’re good at, the problems they can be trusted with, the quiet confidence of knowing they can handle this because they’ve handled it before.

That’s what’s really at stake as the machines take on more of the execution. Everyone is already arguing about the jobs. The quieter loss is the identity underneath them, the sense of being someone who can do something.

And it’s worth being clear about what’s actually eroding, because it isn’t the work itself. Most of us would happily hand over the drudgery. The question is what the drudgery was building in us while we got on with it, and I don’t think many of us ever had cause to ask.

Expertise was a by-product

Here’s the uncomfortable part, and it cuts against the very case I’ve just been making.

The previous piece argued that the call stays ours, that we have to weigh the machine’s confident answer rather than simply swallow it. But that ability doesn’t come from nowhere. You build it by doing the exact work the machine now does for you. You learn to smell a wrong answer because you’ve produced plenty of wrong answers yourself and had to work out why. You develop a sense for what matters by wading through what doesn’t. You know when a confident recommendation has missed the point that actually decides the case, because you’ve missed it before, in front of people, and had to live with it.

The grunt work was the apprenticeship, and nobody ever called it that.

There’s an older model that names this better than I can. The 70:20:10 framework, set out in the 1980s by Morgan McCall, Michael Lombardo and Robert Eichinger at the Center for Creative Leadership, holds that we draw something like 70% of our development from hands-on experience, 20% from other people, and only 10% from formal teaching. You can argue with the exact numbers, and plenty of people do. The shape is what matters. Most of what made you good was never formally taught to you. It came from doing the work, which is exactly the 70% we are now handing to the machines.

It’s worth being precise about how that learning actually happens, because once you see how it works you understand why removing the work costs so much. A great deal of human competence gets built through something that looks a lot like reinforcement. You attempt, you get a signal back, you adjust, and you go again, thousands of times, until the model in your head is any good. Being wrong in front of people, and then having to put it right, is where the learning actually happens; the sting of getting it wrong is the reward signal. We don’t only learn this way, of course. We also learn by being told, by watching others, by the single vivid mistake that never needs repeating. But trial, consequence and adjustment sit close to the centre of how expertise forms, and it’s exactly this loop that handing the work over removes. If the machine produces the answer and you simply wave it through, there’s no attempt and no error to learn from. The signal never arrives.

There’s a sharp irony here. Reinforcement, at enormous scale, is one of the ways the machines themselves are trained: millions of attempts measured against a reward, slowly shaping a better model. So we’re busy industrialising that learning loop for the machines while quietly dismantling our own. We have never been more willing to let something else take the attempts, and the attempts were always where the learning lived.

We’re industrialising that learning loop for the machines while quietly dismantling our own.

Which leaves us a real problem. We need enough expertise to supervise the machine, and that expertise was manufactured by doing the very work the machine has now taken. Automate the production and you starve the faculty you were depending on to do the supervising.

Aviation has wrestled with a version of this for years. Automation has made flying safer and more precise, and nobody in the industry seriously argues otherwise. But the industry bodies and the regulators have also warned, repeatedly, that leaning continuously on automated systems erodes the manual skill underneath. The IATA Aircraft Handling and Manual Flying Skills Report concluded that a significant number of pilots have experienced a degradation of their manual handling skills, and a subsequent over-reliance on automation. Of the pilots who answered the question, more than 90% wanted training to put greater emphasis on the unexpected transition from automatic flight to manual flying: the moment, in other words, when the automation hands the aircraft back. EASA has warned since 2013, in a bulletin it revised again in 2025, that continuous use of automated systems does not contribute to maintaining pilots’ manual flying skills.

The most telling finding is quieter than any of that. The same report notes that pilots rate their own manual flying more highly than instructors’ observed grading actually supports. The skill fades before the confidence does, which is what makes this kind of erosion so hard to spot from the inside.

None of this is an argument against automation. The narrower point is the more unsettling one: a skill you stop practising can desert you at the very moment you need it most.

The rungs everyone is worried about

Nowhere is this sharper than at the bottom of the ladder.

The way most professions have produced experts is by handing the tedious, low-stakes work to the newest people, and letting the expertise accrue as a side effect. The junior lawyer reads the whole disclosure. The graduate does the analysis nobody senior wants to touch, page after page of it, and comes out the other end knowing something that could not have been taught. It was inefficient, and often boring, and it also happened to be how expertise got made. Nobody designed it as a development programme. It simply worked, for free, because doing the work taught the work.

Agents are exceptionally good at exactly that tier of task. So the worry states itself easily: if the rungs are gone, we’ll grow a generation who never climb, and in fifteen years there’ll be nobody with the judgement to supervise anything.

That worry is real, and I don’t want to talk anyone out of it. But it rests on an assumption worth examining, which is that the expertise the next generation needs is the same expertise the last one had.

The craft changed, it did not vanish

I don’t think that assumption survives contact with the work.

If the doing has changed, the competence has changed with it. Someone entering a profession now may never hand-build the thing their predecessors built by hand. What they’ll do instead is harder to name, and easy to undervalue, but it isn’t lesser.

They’ll have to interrogate what the machine produced, and notice where it’s plausibly wrong rather than obviously wrong, which is much the harder catch. They’ll have to know what to ask for and what to leave out, understand the intent well enough to judge whether the output genuinely serves it, and carry responsibility for whatever ships, regardless of what produced it.

That’s a different craft, and in some respects a more demanding one, because spotting a subtle error inside a fluent, confident answer is harder than making the error yourself and being corrected on the spot.

So two things are true at once, and they pull against each other. Some expertise really is substitutable: the mechanical fluency, the syntax, the collation, the sheer grind of assembling what is already known. And some of it almost certainly isn’t, because it’s tacit, the kind that only ever accrued through being wrong in ways you then had to fix yourself.

The trouble is that the second kind always arrived on the back of the first. We never had to teach it deliberately, because doing the work taught it for free.

How does anyone become anyone now?

Which reframes the identity question into something more interesting. Forget “what do I do now the machine does my job?” The deeper puzzle is this: how does anyone become anyone, once the work that used to make us is gone?

If expertise no longer arrives as a by-product of labour, then it has to be cultivated on purpose.

That’s a change of kind. The development that used to happen incidentally, in the doing, now has to be designed, and someone has to do the designing. The ladder built itself for a century. Now it has to be built rung by rung, by hand.

And here the argument turns outward, because none of this can be a solo project. You can’t grow your own judgement in isolation any more than you can be genuinely met by yourself. Judgement grows by making calls in front of someone who can see what you missed, by having your reasoning tested by a person who’s actually paying attention, by being handed work a little beyond you and then not being rescued from it. The tacit passes between people, or it doesn’t pass at all.

So what does that do to the old model? If the 70% that came from doing is being automated away, the 20% that comes from other people has to carry far more of the weight, and carry it on purpose. That is where mentoring and coaching come in, and the difference between them matters more than most people think. Mentoring passes on hard-won experience, the shortcuts and the scars. Coaching does something else: it rebuilds the very thing automation removes, the self-owned learning you only get by working a problem out yourself. I’ve written about that distinction before. Here the point is simply that these relationships now have to do deliberately what the work used to do for free.

So the answer to “who are you when the work changes?” turns out to have two halves. The first is what this series has been building all along: you’re the person who knows what they want, who stays genuinely present to what’s in front of them, and who makes the call and carries it. Direction, connection, judgement.

The second half I didn’t see coming, and it changes the shape of everything that follows. You’re also the person who has to develop that in others. Not as an optional extra, either. Once the incidental apprenticeship disappears, deliberate development is the only thing left that still makes experts, and a senior who simply consumes the output of their juniors and their agents is standing at the top of a ladder with the bottom rungs already gone.

That’s a different kind of work, and most of us were never taught to do it. I don’t have a clean answer for how you learn to grow other people when nobody grew you that way on purpose, and I’m not going to pretend otherwise here. The next piece goes outward, to the leaders who will either build that deliberately or find, a few years from now, that the expertise they were counting on never arrived.

Next Article ->

The Foundation and the Frontier

You can’t hand someone judgement. You can only draw it out of them. AI can bring a newcomer right up to the edge of what a field knows, but only people can grow the judgement to go past it. This piece is about developing others in an age of capable machines.