Last week I sat down with Joe Bonington and Duncan Isaksen-Loxton for their AI Business Flywheel podcast, and towards the end Duncan asked me about the professions: law, accounting, engineering, the places where the junior work was always done by the people coming through. I asked if I could answer it in two parts, because the first part is something that scares me, and I don’t say that often.

Key points

• The junior role was where people learned judgement up close, and AI now does a lot of that junior work.
• Middle management is thinning too, and the middle was where juniors learned from.
• We’ve lost a rung before and rebuilt it. What’s new is losing two at once.
• What’s at risk is the knowledge that never gets written down, and that’s the part people carry.
• It matters in a five-person business as much as in a big firm.
• Three things to start this week: map the learning, keep one rung human, and make room to get things wrong.

The first rung

In June I wrote that we didn’t just automate the jobs, we removed the years. The junior role was never really about the work.

It was where someone learned to read a room, take a correction and recover from a mistake, at close range, with someone whose standards they were learning to carry.

That rung is thinning. A lot of junior work was transactional: moving things around, checking, reconciling, drafting a first version.

That’s exactly the work AI and machines now do quickly and well. Big firms aren’t so much cutting graduates as simply not taking on as many.

MinterEllison, one of the country’s biggest law firms, cut its graduate intake from 104 to 72 for 2025 to 26, and said AI handling more of the routine work graduates start on was part of the reason.

The rung above

And then there’s the rung above it, which I only really saw in the last couple of months.

A manager, as I’d define it, looks after process. They make sure people are doing what they should be doing, and that things are accounted for the way they should be.

A leader is something else. But that management work, the overseeing of process, is also something machines and AI are getting very good at.

So for economic reasons, and now for technological ones as well, middle management is thinning too.

On its own, each of those is a story we can manage. Together they’re a different problem altogether.

Because the entry level was meant to learn from the middle. The middle manager was the one who let the junior make the mistake and then showed them why.

They passed on the intuition, the culture, how the industry actually works and what you do when the textbook runs out. Take both of those rungs out, or even just thin them, and you have to ask where a business gets its ongoing wisdom from now.

Where does it practise? Where does it make its mistakes and learn from them?

Where does it get things right and make a fuss of it?

We’ve been here before, sort of

We have lost a rung before. After the 2008 financial crisis a lot of organisations cut layers of middle management and never quite rebuilt them, and the juniors who came through in the years after found the mentoring they’d expected was thinner than it used to be, because the people left were too stretched to give it.

Go back further and the industrial revolution did something similar to the old apprenticeship, breaking skilled work into steps that could be taught in days, and the crafts that survived did it by reinventing how they passed on what they knew.

So none of this is new in itself. What’s new is the shape of it.

This time the role is still there, the salary is still there, and the work still gets done, often better, but the part of the job that taught someone something has quietly been handed to a machine. And it’s happening at the bottom and in the middle at the same time, which means the people who would have done the teaching are thinning out just as the people who needed teaching stop arriving.

What a business is really losing

Most of what makes someone good at their job never gets written down. It’s knowing that a client who goes quiet is usually unhappy, not busy, when a rule is there to protect someone and when it’s just habit, which supplier will come through on a Friday afternoon and which one won’t, and how to tell someone their work isn’t good enough without losing them.

Nobody puts that in a manual, and nobody could if they tried.

That’s the knowledge that used to travel from the middle to the bottom, one chat and one mistake at a time. AI is very good with the knowledge that’s written down.

It’s the other kind, the kind people carry around in their heads and pass on by working next to each other, that we’re at risk of losing, because the places where it got passed on are the very places we’re making more efficient.

Those things are still important. They sit outside AI.

They’re human needs, and the natural places where they used to happen are not as common as they once were. So where will we let them play out, and who’s going to pay for it?

Why this matters to a small business too

It would be easy to read this as a big-firm problem. It isn’t.

In a business of five people, the owner is the middle manager, and often the only one who knows how things are really done. When I talk to owners about breaking their work down into tasks and asking who each one is best done by, a human, a machine, AI or a mix of the three, the chat nearly always goes well, and the business gets lighter and quicker.

What we don’t always ask is which of those tasks was quietly teaching somebody something.

And there’s a harder question underneath it for any owner thinking about the next five or ten years. If you’re both rungs, the one who knows the work and the one who checks it, then the business’s judgement lives in one head.

If the juniors you’d normally have trained up never come through because the tools now do their work, who’s going to be ready to run it when you step back, sell it or hand it on?

The ripples

Follow it out a few years. Firms that hired fewer juniors find, later on, that they have fewer people ready to step up, because nobody did the years.

Experienced people retire with their judgement still in their heads. Clients get answers that are fast and correct and somehow thinner, because nobody in the room has seen the thing go wrong before.

And the people who did come through with good judgement become very valuable indeed.

Then there’s the ripple after that one. The mentors of the 2030s will be the people coming through right now, and if their early years were spent checking a machine’s work instead of doing it themselves, they’ll pass on a different set of instincts to the people after them.

Not necessarily worse ones, but different, and nobody will have decided that on purpose.

This isn’t an argument against AI, mind. I’ve built it into everything I do over the last few years, and it’s given me back the part of my work I love.

It’s an argument for deciding on purpose where the learning goes, so we don’t find out in ten years that nobody taught anybody.

Three things you can do now

I don’t have the whole answer yet, and the world isn’t waiting on me for it. But here are three things any business can start on this week, whatever its size.

First, I’d map the learning as well as the tasks. When you sit down to work out who or what does each task, add one more column: does this task teach someone something?

If it does, think twice before you hand all of it to a machine, or keep a person close to it on purpose. You’ll usually find it’s only a handful of tasks, and they’re often the slightly messy ones nobody likes.

Second, keep one rung human, on purpose. Pair someone junior with someone experienced on real work, not training work, and treat that time as an investment, the way the old firms did.

The inefficiency was always the investment. If you’re a small business, that might be one afternoon a week where the person who knows sits beside the person who’s learning and lets them do it, mistakes and all.

And third, make room to get things wrong. Decide where your people can try something, get it wrong where it costs little, and talk it through.

If the natural places have gone, build one, and make a fuss of them when they get it right. A short weekly “what went wrong and what did we learn” over coffee does more than most training courses ever did.

The question I’m left with

On the podcast I said I’m not sure what the resolution is, and that’s still true. Joe, Duncan and I went further into what all this means for experts and for the professions, and I’ll share that when the episode’s out. In the meantime I’d love to know how you’re handling it. Who’s teaching your juniors now, and who taught the people teaching them?

Choose forward.

If you’d like a short read every Tuesday on what’s coming and what it means for your work: welcome.morrismisel.com/join

---

Morris Misel is a foresight strategist, keynote speaker, and trusted adviser. He works with leaders, organisations, and boards to interpret what’s changing and make better decisions about what comes next. Heard by more than a million people a year on stage and in the media.

---

Questions people ask

Is AI cutting graduate and entry-level jobs?

In some places, yes. MinterEllison cut its graduate intake from 104 to 72 for 2025 to 26 and said AI was part of the reason. The bigger question is what juniors were learning in those roles, and where they learn it now.

Why does middle management matter for learning?

The middle manager was the one who let a junior make a mistake and then showed them why. They passed on the intuition, the culture and how the work is really done.

Isn’t this just another round of automation, like the ones before?

Partly. We’ve lost a rung before, after the 2008 crisis and much earlier with the old apprenticeships. What’s different now is that the bottom and the middle are thinning at the same time, so the people who’d do the teaching and the people who’d be taught are both getting scarcer.

What kind of knowledge is at risk?

The kind nobody writes down: reading a client, knowing when a rule matters, judging a supplier, giving hard feedback well. AI is good with written knowledge. This sort is passed on by people working side by side.

Will experienced people become less valuable?

If anything, the opposite. If fewer people come through with real judgement, the ones who have it become worth more, which is a good reason to make sure they’re passing it on.

What can a small business do about it?

Add a column to your task list asking whether each task teaches someone something. Pair someone junior with someone experienced on real work. And set aside a place where people can get things wrong where it costs little, and talk it through.

Sources

• Lawyers Weekly, Grace Robbie, 20 May 2026: AI is coming for BigLaw grads: what shrinking intakes mean for Australia’s future lawyers.
• AI Business Flywheel podcast with Joe Bonington and Duncan Isaksen-Loxton, recorded 8 October 2026 (episode link to come).