The first job on a shoot is rarely glamorous. You’re the runner. You’re in before the crew, you know everyone’s coffee order by 10am, and you spend most of the day carrying things from one end of a location to the other. But that was never really the job. The job was standing three feet from the DoP while they argued about a lens. It was sitting at the back of the edit at midnight, watching someone cut the four seconds that made the whole spot work. Nobody wrote any of it down. You just absorbed it.
That’s how most of the craft in this industry has been passed on: by osmosis (and an awful lot of coffee). Which is why one line keeps coming back to me. AI won’t take your job, but it’ll probably take the task that trained you. That matters for all of us, and for the colleagues we haven’t hired yet.
The groundwork – the first-pass edit, the resizes, the research deck, the transcript, the 40 headline options is now work AI does faster and cheaper. For an agency or production company under margin pressure, the rational move looks obvious: stop hiring juniors and let the mid-weights prompt their way through it.
I understand the logic. We’re close to the point where there’s no good economic reason to pay a young person to do work AI can already do. But I’d argue we must, because the alternative is an industry that stops growing its own people. If nobody runs, nobody learns how a set works. If nobody sits in the edit, nobody becomes an editor. And in 10 years we’ll be asking where all the mid-weights went.
The shape of our organisations are already changing. We’re moving from a pyramid to a diamond: a smaller bottom, a broader middle and a top that stays roughly where it is. That isn’t necessarily a disaster. A smaller junior intake can still be a better-trained one, but only if someone decides to train them.
That’s where it gets awkward. With so much of the market now freelance, who actually owns that? Freelancers can’t afford to. The agency assumes the production company will. The production company assumes the agency will. And the junior is left learning from a chatbot at 1am, which teaches them a lot about answers and very little about taste.
Here’s the bit I think most of us are missing. The juniors coming in now have a genuine advantage. They’re building AI fluency from day one, with no old habits to unlearn. I’ve had Excel on my desktop for three decades and I still can’t build a pivot table. The 22-year-old on your team will have automated hers before lunch.
The choice isn’t AI or juniors – that’s a false dichotomy. Plenty of the seniors expected to replace juniors with AI haven’t been properly trained to use it either. Look at Brainlabs, the independent media agency of around 1,000 people. It has always run an academy for entry-level hires, and since October 2023 it has taught each cohort to work with AI from day one rather than bolting it on later. The intake went from 19 in that first cohort to 64 in April 2026, a 237% increase. Founder and global CEO Dan Gilbert told Forbes it didn’t fund this by cutting junior roles to pay for tools. Juniors trained in AI simply start adding value sooner, so the agency can take on more clients or do better work for the ones it has. Same technology. Different decision.
If we want that advantage to become talent, it takes three deliberate choices.
-
Structured learning, not hope. Bring back the grad scheme, the apprenticeship, the simulated brief where a junior can do a version of the real work and get it wrong safely.
-
Visible modelling. If the craft now lives partly in how you prompt, review and reject what the machine gives you, seniors have to do that out loud, where juniors can see it. Osmosis needs proximity.
-
Changing what gets measured. If utilisation is the only number that matters, nobody will spend an hour teaching. So count the teaching.
None of this is free. The government’s new youth jobs money helps at the edges, but a training grant doesn’t pay a London salary. Someone still has to decide that the role should exist.
So pick one junior. Put them in the room where the real decisions get made. Show them how you use the machine, and when you ignore it.
Who trained you? And who are you training?