A recent hire told me the thing she was most nervous about wasn't that AI would take her job someday. It was that it had already taken the job she was supposed to have first. The tasks that used to be handed to the youngest person in the room — clean up the deck, reconcile the numbers, draft the memo nobody would read closely, sit in on the call and take notes — were the ones a model now does in seconds. She wasn't wrong to notice it. What she was describing is a gap opening up quietly at the bottom of nearly every profession, and I don't think we've been honest about the cost.
Here is the part that's easy to miss: that entry-level work was never really about the output. Nobody needed the notes from that call badly enough to justify a salaried person taking them. The reconciliation could have been wrong and someone senior would have caught it. The point of the grunt work was the reps. You learned what a good deck looked like by making bad ones. You learned where the numbers hide by being the person who had to find them. You absorbed judgment sideways — from being close to decisions you weren't yet trusted to make.
The bottom rung was never about the output
When you remove the grunt work, you don't just make juniors more productive. You remove the curriculum. We are now asking people to start where they used to finish — editing instead of drafting, judging instead of building, approving instead of doing. And editing is a skill that sits downstream of making. You cannot reliably tell whether a model's first draft is good if you've never had to write a bad one yourself and feel where it went wrong. The judgment we're asking juniors to apply on day one is exactly the judgment the old path spent years quietly installing.
You can't skip the reps
What worries me as an employer isn't this year. It's the pipeline. If the math says a model produces entry-level output cheaper than a junior hire, the rational short-term move is to stop hiring juniors. Do that across an industry and you're eating the seed corn. Ten years from now there is no deep bench of mid-career people, because nobody was allowed to be junior long enough to become one. Everyone ends up bidding for the same scarce pool of senior talent that happened to come up before the door closed. We'll have optimized the current quarter by quietly defunding the next decade.
At Provieo we've started treating this as a design problem rather than a hope. The instinct is to give a junior person the work a model can't do yet — but that work is usually too big and too consequential to hand to someone still learning. So instead we deliberately carve out small, real things they can own: decisions with genuine consequences, sized so that being wrong is survivable. Not busywork the model could do, and not a stretch assignment that fails loudly. The goal is to manufacture, on purpose, the reps that used to happen by accident — because the accidents that produced them aren't happening anymore.
Teaching has to become deliberate
That changes the manager's job more than the junior's. Mentorship used to be a byproduct of proximity. You learned by being in the room, and the teaching happened whether or not anyone intended it. Take away the shared grunt work and the proximity goes with it, and the learning that used to be free now has to be built and paid for with someone's attention. The managers who understand this are going to spend real time engineering the reps their people no longer stumble into. The ones who don't will wake up in a few years wondering why nobody on the team can be trusted with anything hard.
I don't think the answer is to romanticize tedium or to slow down and do things the hard way out of nostalgia. The tools are good and they're not going back in the box. But a company is a place where people are supposed to get better at their work over time, and that only happens if the ladder still has a bottom rung. Rebuilding one is slower and less efficient than letting the model do everything a beginner would have done. It's also the cheapest talent strategy I know of — you just don't get the return for a decade. The firms willing to wait that long are going to have a bench when everyone else is fighting over the same handful of people who got in before the door swung shut.