We automated our apprenticeship. Now someone has to rebuild it

The industry has decided that the rise of AI is a jobs story. How many juniors and which roles will be impacted. What the ladder looks like now. I believe this is the wrong angle. This is a training story, and looking at it any other way will cost us.

For decades, people learned this craft through work that senior leadership agreed was beneath them. Forty versions of a line, rolling someone else's idea out across fourteen formats, rebuilding a deck in the early hours. You learned what good was by producing an enormous amount of not-good and being told why. Not glamorous, but this was the curriculum.

The economist David Autor named this Polanyi's paradox, after the philosopher who wrote in 1966 that we can know more than we can tell. The hardest work to automate, Autor pointed out, was always the work people could do but not fully explain.

That's judgment. The call you make in a second and could not write down if you tried. That's the safest thing a creative person can own, and why those menial tasks mattered, and matter.

Machines haven't taken the judgment. They've taken the grunt work that used to teach it. Removing the mechanism that helped people build that unexplainable skill and changing nothing else doesn't get you a leaner industry. It gets you, in a few years' time, a layer of senior people who never developed a standard, sitting in rooms approving average work produced by AI because they have no basis on which to object.

That's not a talent shortage, it's a judgement shortage that will show up in the work long before anyone takes notice.

The good news is that the curriculum was never really the tasks. It was the feedback loop wrapped around them, which can be rebuilt, even faster than the original, which was hopelessly inefficient anyway. Nobody needs three years of resizing to learn what balance is.

Four simple approaches can change the narrative, if the industry is willing to put the effort in.

Make rejection the exercise: Hand a junior the machine's output, all seven options. Make them kill six with reasons a client would accept, then argue with those reasons. That is the single most useful hour you can spend on someone in their first two years, and the raw material is free to produce.

Run the brief backwards: Give them a finished piece of work and ask them to write the brief it answers. Much of the confusion a junior encounters is not being able to see the problem underneath the solution, not the craft.

Measure decisions, not output: I have never seen an entry-level review that did not count what someone made. Volume of production has stopped being a meaningful measure of a person. The volume and quality of their decisions has not.

Get the senior people talking: Ask a good creative director why they killed something and you will hear that it did not feel right, which is both true and useless. The reasoning is real, but dragging it into words takes patience and repeated questioning. It will feel to them like being asked to explain a reflex, but it is still the most valuable thing they can give away.

These approaches aren't free. There's a real cost in terms of time and they produce nothing billable this quarter. The old apprenticeship was cheap because it hid inside work the client was already paying for. This one has to be paid for in a different, almost foreign way. Which makes it a leadership decision, and at most shops right now it is being made by default.

Nobody has decided to stop developing people. They have decided not to start. We pulled the ladder up because the bottom rungs stopped being worth the cost. That's fair enough. It was never a very good ladder anyway. But somebody has to build the thing that replaces it, purposefully. It will not build itself while the whole industry stands around waiting to see what the machines do next.

Marc Lewis is the dean of the School of Communication Arts and founder of SCAFFOLD AI.