Expert hands, junior judgment

Most conversations about implementing AI start in the wrong place. People treat it like buying a tool — a licence, a box, a feature you switch on. In practice it feels a lot more like making a hire. Give an agent the right tools and you haven’t really bought software; you’ve brought a new colleague on board. And once you look at it that way, you can build out a whole digital team — someone for marketing, someone for sales, for infrastructure, for development — each one a teammate you brief, supervise, and correct. The only strange thing is that the colleague you just hired comes with a very lopsided CV.

A very lopsided CV

Give it the right tools and, at the actual task, it’s an expert. It writes the config, drafts the email, refactors the function, queries the API — quickly, tirelessly, and often better than the people around it. Ask it about the field it’s working in and it settles into something more like a solid mid-weight: it knows the textbook cold, far broader than most people you’ll meet, but it’s thin on everything that never got written down — the politics at the client, last year’s outage, why nobody touches that one server on a Friday. And the moment a real decision is on the table, it’s a junior again — confident, plausible, and quietly in need of someone to look the work over before it goes out the door.

You’ve never actually hired anyone who looks like that. Senior at the keyboard, junior in the room.

Why the judgment stays human

That last gap, the one around decisions, isn’t something a bigger model quietly closes. It runs deeper than raw capability. An agent has no skin in the game: it doesn’t live with the consequences of being wrong, it can’t be held to account, and it won’t be in the room for the post-mortem. Judgment is more than spotting the pattern — it’s being the person who owns how things turn out. Wearing my other hat, the IT-Revisor one, the point is hard to dodge: someone has to carry the responsibility, and a model simply can’t.

The rest of it keeps improving. Execution gets sharper with every release, and so does the sheer breadth of what these systems know. But the question of who answers for it when something breaks still lands on a person — by design, and more and more by law — and that doesn’t shift with the next model. So even your fastest, sharpest agent still needs a human standing behind its decisions, which is really just what a good senior has always done for a junior.

Where the seniors come from

The popular conclusion runs into trouble right about here. It tends to go: agents handle the repetitive, entry-level work, so we can stop hiring for it. The first half is fair enough — that routine work really is what an agent does best. It’s the second half that falls apart.

Seniors aren’t bought in, they’re grown, and they grow out of being juniors: doing exactly that repetitive work, getting it wrong, and slowly learning why it was wrong in the first place. Take away the jobs people start in, across a whole industry, and you quietly remove the only road anyone had to experience. Each company trimming its juniors looks perfectly sensible on its own books, while together they saw off the branch the next round of seniors was supposed to grow on.

A different kind of junior

So the answer probably isn’t to replace juniors at all, but to change what the job actually is — and to do it early. The junior of the next few years can start much closer to where we used to put someone mid-career: not churning out the work, but supervising it, framing the problem, judging whether the result is any good, and knowing the field well enough to catch the agent when it’s confidently wrong. Less time on the busywork, more on the thinking, and sooner. That’s a different person than the industry hires and trains for today, which is why the real change sits upstream of any one company — in how we teach people and bring them into work. That’s a bigger lever than I’ll ever get to pull. The smaller version, though, is already on my desk: hire for judgment, and trust a junior with the supervisor’s seat a little before they feel ready for it.

You can buy the expert hands. You still have to grow the judgment.

The firms that come out of the next few years ahead won’t be the ones that cut their juniors the fastest. They’ll be the ones that turned those juniors into supervisors from the start — and held on to the first job everyone else was so eager to automate away.