There’s a comfortable consensus worth revisiting: that every technological wave belongs to the young, and the rest of us are left running to catch up. With generative AI, the opposite is happening, and almost nobody says so out loud. Today’s models are a machine for amplifying professionals with decades of experience.
The reason is less mysterious than it looks. A model writes code, drafts, summarises and proposes architectures at a speed no team can match. What it doesn’t do is know which of those proposals will survive contact with a specific client, a specific budget and an eight-year-old legacy system. That judgment comes from the person in front of the screen, and it comes reliably only if they’ve accumulated enough hits and misses. Put more directly: AI multiplies judgment, and judgment can’t be downloaded — it’s built over years.
AI multiplies judgment, and judgment can't be downloaded: it's built over years.
What the data says, not the keynotes
This is no longer an industry hunch. The Stanford Digital Economy Lab has spent three years analysing the payroll records of millions of US workers through ADP, the country’s largest payroll processor. Its conclusions are consistent: since late 2022, employment among 22-to-25-year-olds in the occupations most exposed to AI (software development, customer service, administrative roles) has fallen significantly, while employment among experienced workers in those same roles has held steady or grown. The gap between the two now exceeds 19% and keeps widening.
The researchers’ explanation matches what we see every day: AI replicates codified knowledge well — the kind that comes from training and books — and replicates tacit knowledge badly — the kind only years of practice can give. Put more bluntly: AI knows what a brilliant new graduate knows. What it doesn’t know is what your best senior knows.
Companies have reacted fast. According to SignalFire, new graduates have gone from 15% of Big Tech hires before the pandemic to barely 7%. At the same time, executives at the major AI labs are announcing that their models will absorb a large share of entry-level jobs. Those announcements are worth reading for what they are: messaging aimed at investors, funding rounds and IPOs. They are not a strategic plan for your company, even if they’re being treated as one.
Two independent measurements of the same shift: the entry rung is narrowing while the top one holds.
19%
is the employment gap that now separates 22-to-25-year-olds from veterans in the occupations most exposed to AI, and it keeps widening
Stanford Digital Economy Lab · ADP payroll data, 2026
7%
of Big Tech hires are new graduates; before the pandemic it was 15%
SignalFire · State of Tech Talent, 2026
It isn't AI taking work away from everyone equally: it's the first rung narrowing.
“But AI already does that job”
It’s the logical objection, and it’s partly right: a good share of the tasks a junior profile traditionally took on are now handled by a model. If the analysis stops there, the conclusion is obvious and wrong.
Because a junior was never just someone who does low-cost tasks. They’re the only factory of seniors there is. Nobody arrives on the market with judgment: it’s built by giving responsibility to people without experience, letting them make mistakes in controlled settings, and sitting them next to someone who already made those mistakes. That process takes years and admits no shortcuts. The company that decides today that a software subscription should fill the entry rung is signing up to have no forty-year-old seniors in 2035 who are trained in its culture, its systems and its clients. In tech, we’ve had a name for that pattern for decades: technical debt. This is the version with people, and the interest is considerably worse.
There’s also an asymmetry no model can resolve. Every project handed to an AI starts from zero: it doesn’t accumulate the organisation’s context, doesn’t remember why a decision was made, doesn’t carry the consequences of anything. A young person does. They accumulate, connect, commit. And they bring something veteran teams lose without noticing: a fresh eye. Well supported, today’s junior profiles will be the first generation to grow professionally with these tools built in from day one. Giving that up to save an entry-level salary is optimising the quarter at the expense of the next decade.
Automating or amplifying: a decision, not a destiny
The most important nuance in the Stanford study is the one that generates the fewest headlines: AI’s impact on youth employment isn’t uniform. Where AI is used to automate tasks, entry-level employment falls. Where it’s used to increase people’s capacity, it holds steady or even grows. This isn’t a law of physics for technology. It’s an organisational design decision, and that decision gets made in a leadership committee, not a data centre.
Meanwhile, at the other end of the board, China’s big tech companies are making the opposite bet: recruiting talent directly out of high school, with contracts signed before university, and putting chief scientists under 35 in charge of their strategic AI and robotics projects. It isn’t a model to copy wholesale, and it carries costs we wouldn’t want to import. But the direction is unambiguous: there, young talent is fought over as a strategic asset; here, it’s cut as a cost line. Only one of the two approaches is thinking about 2035.
Where we stand
At Digital Bakers we work with AI every day: agents, automation, pipelines. Short term, we’d do fine selling the fashionable pitch — automate everything, cut teams. Our position is different, and not out of nostalgia but out of common sense: AI is the best thing to happen to our industry in twenty years, provided it’s used to amplify people, not to amputate the base of the organisation.
So if you’re deciding what role AI will play in your company, the useful question isn’t whether a model can do a junior’s job. It can — that’s what it was trained for. The useful question is where your next senior is going to come from. The companies that can answer that will have, in ten years, something no subscription can buy: people with judgment, homegrown, who master these tools better than anyone.
Julio Prieto, CEO of Digital Bakers
Sources
- Stanford Digital Economy Lab (Brynjolfsson, Chandar, Chen), Canaries in the Coal Mine? — ADP payroll data and youth employment in AI-exposed occupations, 2026 update.
- SignalFire, State of Tech Talent Report, 2025 and 2026.
- Coverage of youth talent recruitment at Chinese tech companies (Xataka, Sixth Tone, Le Grand Continent).