The data on early-career hiring is genuinely bad: a Stanford study found employment for 22-to-25-year-olds in AI-exposed fields now sitting 19% below where it would be had it kept pace with less-exposed peers since late 2022. But that number is being read as “AI is deleting the entry-level rung,” and that’s not quite what Data Society’s Global VP of People, Catie Maillard, is seeing. Her read is that the roles aren’t disappearing so much as being rebuilt around different work. The admin work, data entry, and process legwork that used to double as on-the-job training is exactly the work AI still needs a human checking, which means entry-level hiring isn’t disappearing so much as getting redefined around it.
The headline stat is real. Stanford researchers, including economist Erik Brynjolfsson, found that employment for early-career workers in the most AI-exposed occupations, software engineering, marketing, and customer service among them, fell 19% relative to older workers in the same fields since ChatGPT went mainstream in late 2022. Workers 35 to 49 in those same fields saw employment grow by about 10% over the same period, while experienced workers overall showed no comparable gap between more- and less-exposed occupations.
The study’s own explanation for the split matters as much as the number: roles where AI automates a task outright lost ground, while roles where AI augments a person, helping them learn or checking their work, held up. That distinction is close to the one Maillard was making when she was asked directly whether AI is cutting the roles entry-level workers used to use to earn their stripes and build a ladder.
“The question was around whether AI is cutting the roles that entry-level folks used to come in on, earn their stripes, and figure things out, and how do we build a future ladder. I do feel like that’s a really hard question to solve for, because we’ve had folks come in handling a lot of the admin tasks, and in doing so were learning the larger processes, learning the larger context. Even from a finance perspective: someone going in and double-checking the ledger, double-checking spreadsheets, working on data entry. While that’s not necessarily work that requires you to have experience, you now have better context that you’d be missing if you went right into, let’s say, an accountant role versus an accounting assistant role.”
Catie Maillard, Global VP of People, Data Society
Maillard’s answer doesn’t dispute the Stanford data. It reframes what to do about it. Her position is that entry-level roles aren’t obsolete. The exact thing AI is worst at, working from messy, incomplete, or wrong data, is exactly what entry-level and admin work has always caught.
“So my feedback on this, and at least what I’m seeing, is that for the most part I still see a need for entry-level roles. It’s just what they’re working on is changing. I think we still have issues with data quality, because the AI works best when we are feeding it clean data and when it can spit back out clean data, and there’s still a need for folks to be looking and reviewing and doing some of that admin work, if not more than ever. I found this specifically in HR. My need moved less to employee-facing and more toward back-office needs, because I’m able to be the employee-facing person more often now, since AI can enable me here. It can do a lot of the tasks I need, but it only works if I have all the data cleaned in our HR system, if our processes are kept up to date, and a lot of that work is actually more on the entry-level or admin side. So I’m answering this question by challenging whether or not entry-level roles really need to go away, particularly in the world of AI. I feel like I need them more than ever. I’m seeing more mid-level roles be more of a ‘hey, can I wait and see if I can maybe hire someone less expensive who, plus AI, can get things through.’”
Catie Maillard, Global VP of People, Data Society
“Garbage in, garbage out” is old advice in data management, but it’s become the operating constraint on every AI deployment: a model built on clean, current, well-structured data produces usable output, and one built on messy data doesn’t, no matter how capable the model is. Maillard’s HR example is a small, specific version of that constraint. AI can free her up to spend more time with employees, but only if someone is still keeping the HR system clean and the processes current, and that someone is, by her account, more likely to be entry-level or admin staff than ever.
The part of Maillard’s answer worth sitting with is the last line: the squeeze isn’t hitting entry-level hiring so much as mid-level hiring, where managers are increasingly asking whether one person plus AI can cover work that used to take a full mid-level hire. That lines up with the Stanford finding that automation-type roles are losing ground while augmentation-type roles are holding up. If a mid-level job is mostly a person doing a task AI can now do end to end, that job is exposed. If an entry-level job is mostly a person catching what AI gets wrong, building the context that makes later judgment calls possible, that job still has a reason to exist, and by Maillard’s account, more of a reason than it used to.
Before cutting an entry-level role, ask what it actually produces: is it work AI can fully automate, or work where a person catches what AI gets wrong? Data Society’s AI upskilling programs help teams redesign entry-level roles around that second category on purpose, and AI Advisory can help map where your own pipeline is most exposed.
