In brief. Automating entry-level tasks can take away a junior’s chance to learn. The first assignment has to be designed so that the new hire practices deciding. At Anthropic the share of juniors among new technical hires fell from 27% to 9% (study of October 16, 2026).
“It’s nice to see people your age discovering these things.”
A young man fresh out of university said it to me at the end of a two-day bootcamp on bringing AI into a company. About my age, nothing to add. I acknowledge the generation gap. It was the verb that stung: discovering. I had spent two days explaining where that technology was heading and, in his reading, we were all happily busy discovering it.
We were, by the way, at the basics. He had not just built a model or switched on a home server with thousands of agents at work. But he was familiar enough with the tool to place us, rather generously, in the category of latecomers.
The remark annoyed me. Then it seemed useful. In that room we were using two different yardsticks to measure a skill. For him, at least in that moment, what counted was already knowing the technology. I was trying to understand what we would do with it inside an organization, where a wrong answer would have a cost.
The work that taught you how to work
In the study published on October 16, 2026, Anthropic reports that the share of juniors among new technical hires fell from 27% to 9%. Their presence in technical departments also shrank, from 21% to 14% since the start of 2025. The company keeps growing: those percentages describe the composition of the workforce, and do not by themselves show how many jobs AI has eliminated. Anthropic’s original study, October 16
The figure interests me because it puts pressure on a fairly elementary mechanism. You joined a company, you were given relatively simple tasks and, by doing them, you learned how the more complex work functioned. The first comparison of quotes produced a table. It also taught you that two offers with the same price could hide very different conditions.
Now the table can arrive ready-made. And along with the effort, the chance to ask why an experienced colleague discarded the cheapest offer can disappear too. A transport cost, a delay seen before, a condition in the small print: judgment also forms by running into details we did not know to look for.
I have no nostalgia for days spent copying data. The objection to this reasoning is fair: making a young person do pointless work just because we did it is not training. It is a rite of passage that risks becoming a punishment. AI can free them sooner, let them try more solutions and take on problems that used to be out of reach.
That advantage, though, has to be organized. Leaving a person in front of the correct result is not the same as teaching them to recognize it. The first assignment must contain a decision that belongs to the new hire, even when most of the execution is automated.
Before opening the answer
In a purchasing office you can take a choice that has already been made, remove the outcome from the file and ask the junior to justify a recommendation. Which supplier do they choose? What information is missing? What would make them change their mind? Only then do you open the agent’s work and compare the two paths.
This is an exercise I am suggesting, not a success story to show off. It works if the file contains a real difficulty and if someone is able to discuss the answer. A comparison between two nearly identical quotes teaches little. An apparently convenient offer with a hidden production constraint forces you to understand the trade.
In the Scenari episode on studying in the age of AI I talked about the value of using these tools to explore a line of reasoning. In a company I would add one condition: when the comparison happens, the new hire has to bring a hypothesis of their own. It can be wrong. That is exactly what makes visible where they went astray. Scenari · Studying in the age of AI (in Italian)
Responsibility can grow without putting a customer at risk. First you decide on closed cases, then on a real file with review, finally within an autonomous perimeter. The step up depends on what the person is able to justify. The speed with which they use the tool can help, but by itself it does not tell you whether they are ready to answer for the consequences of their choice.
The experienced colleague is asked for a different job from correcting the sheet. They have to explain a step that has become automatic for them. It is often the hardest part: years of experience compressed into “I’d leave that supplier alone.” The young person needs to know what is inside that sentence, so they can recognize the next case without picking up the phone every time.
This time has to be assigned. If the senior has to train while absorbing all the work the agents produce, the comparison will be the first activity to be dropped. The department plan therefore needs two dates: when the agent takes over the repetitive task, and when the new hire must be able to make the first decision on their own.
In the end, I would like that young man to be right to feel ahead. I would just like to measure that advantage on the first problem he can solve, explain and defend in front of a customer. At that point he can make all the jokes he wants about my age.
Read next One department’s productivity, another’s queue
Background Impact · the case file (in Italian)
Image: Anthropic