Executive summary
Simply put, you can’t remove rungs on the employment ladder without compromising the structure of the ladder as a whole.
With entry-level job postings down double digits since the launch of ChatGPT, AI’s impact on the workforce is felt most acutely in entry-level jobs. Still, everyone is spending money on and using AI. Already, 88% of businesses now use AI for at least one business function, up from 33% two years ago. Meanwhile, AI infrastructure spending last year matched the size of the entire global venture capital market in 2024. The thing is, we can’t see an economy-wide AI shock. In fact, overall US employment is up 6% since ChatGPT’s release.
Most studies and research look at the broader job market and see nothing. That’s because AI's impact on the workforce isn’t evenly distributed.
The reason the impact of AI is so hard to spot comes down to basic maths: the people most affected, employees aged 22 to 25, make up fewer than 1 in 10 employees. A recent study from the Stanford Digital Economy Lab, Canaries in the Coal Mine?, addresses this issue, indexing employment to 100 at the launch of ChatGPT and tracking how this changes over time. Their measure of AI exposure ranks occupations by how much GenAI (generative AI) models can currently do in aspects of their daily work.
Results use a five-year balanced sample of firms using ADP payroll services and include all workers matched to an occupation code.
Split the data by age and by AI exposure, and you begin to see a pattern. Employment for early-career employees, 22 to 25: down by an average of 11% in the most AI-exposed occupation quintile (the fifth of employees most exposed to AI) between November 2022 and June 2026, 19% below the least exposed quintile. Employment for employees aged 35 to 49 in those same quintiles rose by about 10% over the same period, and the inter-quintile gap was no longer statistically significant. In short: demand for entry-level jobs is shrinking while demand for mid-to-senior talent grows.
Results use a five-year balanced sample of firms using ADP payroll services and include all workers matched to an occupation code.
In an August 2026 update, the study expanded to cover 4.6 million employees across more than 730 occupations. They reached the same results and conclusions, even after excluding technology firms and computer-adjacent jobs, roles capable of being made remote, and interest-rate exposure. This pushes back on research from the New York Fed, which attributed most hiring slowdowns to overhiring during the pandemic and high interest rates.
This isn't unique to the USA. Economists from the Dutch bank Rabobank reached the same conclusion with Dutch data. Between the fourth quarter of 2022 and the third quarter of 2025, employment among 15- to 24-year-olds in the most GenAI-exposed occupations fell by more than 13%, while employment in other occupations rose 3%. At the same time, vacancies for those jobs dropped by 25%, a loss of 19,000 job openings. A Bank of Korea study on South Korean employment also reached similar conclusions. Different independent studies using different methodologies arriving at the same conclusion strongly indicate that this is structural and not just a normal part of the hiring cycle.
Canaries in the Coal Mine? states this mechanism directly:
“Why might AI disproportionately affect junior workers? One possibility is that AI more effectively substitutes for knowledge that has been digitized and codified—the core of formal education.”
They connect it explicitly to age:
“AI may be automating the checkable, process-intensive tasks that historically justified junior headcount, while increasing the leverage of experienced staff." In other words, AI hits entry-level jobs harder because it can automate “entry-level” tasks while increasing the productivity of seniors.
Results use a five-year balanced sample of firms using ADP payroll services and include all workers matched to an occupation code.
The tasks involved in entry-level or senior jobs split along a codified-tacit knowledge line. Codified knowledge is the stuff you can write into a rule, a script, or a training document and hand to someone else to execute. One key distinction between AI and human understanding is tacit knowledge. Employees build tacit knowledge cumulatively by actually doing a job. It’s something that isn’t going to show up in a model’s training data. At ML6 it’s the difference between executing a code migration strategy and designing it; the difference between doing a task and managing a project. That split predates AI by far longer than twenty years.
Here’s the thing: that’s the economic logic behind careers—you start with documented procedure and tasks with a checkable answer. Those are the tasks a person can do competently before judgment has had time to accumulate, and doing them builds that judgment. A senior employee’s day is what comes after that. It deals with the exceptions where checklists become useless.
Industry leaders agree that a shortfall in entry-level hiring is a problem. Cutting the entry-level rung will result in an employment “succession crisis.” Without staff building expertise at the entry-level and developing a ‘gut feeling,’ you’ll burn through your senior talent and find yourself with nobody qualified nor capable of handling senior management decisions properly.
This isn’t some far off, opaque eventuality. If a firm doesn’t hire an entry-level class in 2026, there’s a shortfall in middle management by 2029, and senior leadership in the following decade. Before you know it, you’ve crippled your entire talent pipeline for the sake of efficiency and cost-cutting.
Simply put, you can’t remove rungs on the employment ladder without compromising the structure of the ladder as a whole.
That's just on the company level. You can ask yourself: if every company follows this trend, could the lost jobs across society also lead to higher unemployment rates for those seeking entry-level jobs? And in turn, how might that affect consumer spending, demand, and a potential cycle of pressure to cut costs even more? Every firm’s headcount saving is some other firm’s missing customer.
Entry-level employees will need to differentiate themselves with soft skills and critical thinking, which are harder for AI to learn. A trend we've seen at our own company is a shift in engineering roles, for example. Where a couple of years ago, ML6 focused heavily on classic machine learning expertise, we currently hire more (GenAI) engineers who can also build the bridge towards production-ready solutions.
As AI automates the routine and repetitive tasks we associate with entry-level jobs, the value an employee brings to the table is their human-centric soft skills and their experience. By definition, entry-level employees do not have the latter, which means focusing on soft skills is key for people trying to climb the career ladder.
Critical thinking, people management, team building, project management, these are all examples where having technical knowledge is useful but having the ability to apply that knowledge laterally and deploy it in novel ways is invaluable. It’s what differentiates a regular, qualified, candidate from a versatile one.
The hiring manager approach shouldn’t be too different from entry-level employees. Hiring managers need to actively think about the fact that senior skills are learned at the entry-level, and if they automate these jobs, where will their next generation of senior and middle management come from?
The best way to protect against this is to be clear about what entry-level employees bring to the table, specifically what they bring to the table regarding AI. In our own teams, we consistently see younger hires adopt new AI tools the fastest. They explore and discover new ways of using AI that more senior employees wouldn’t instinctively see. The value of this as an AI tooling resource, whether internal or external, is immeasurable. So we say, automate the repetitive tasks, the routine tasks, but do not automate the tasks which teach judgement, critical thinking, and other soft skills and definitely do not automate entire roles out of existence. If nobody is learning your senior skills internally, you’ll have to find them externally. But if everyone is automating the jobs where those skills are learned, the competition for these skills will be pretty rough.
The advantages from AI are larger than just productivity gains. Protect your talent pipeline by building AI literacy among your entry-level staff; this allows you to easily redeploy talent laterally across the company and reduces the chances that hires become AI-replaceable; do this by augmenting their work with AI instead of automating it. It balances the productivity gains from AI with a human touch. It gives you the opportunity to redesign entry-level jobs around judgement and soft skills instead of box-ticking which should further legitimise the product/service you are offering. Nobody wants to hire a company where the product is just an LLM (large language model) behind a fancy facade. Through this, you both give yourself a reason to maintain entry-level headcount and lay a strong epistemological foundation for future mid-to-senior management.
Just like entry-level employees, hiring managers should focus on skillsets instead of knowledge. Future hiring will likely become more skill-based instead of role-based, as AI increasingly takes over individual tasks. Redesigning entry-level hiring around this principle and focusing more on skills that demonstrate adaptability makes your hiring pool more versatile, reducing the risk of AI-induced redundancy and therefore a greater ROI on the time invested into a new hire. Then double down by investing in building your employees’ skillsets through AI-literacy courses and upskilling. If you’re interested in the topic, the shift from role- to skill-based hiring is something our Talent & Culture team recently discussed.
At ML6, we support our entry-level employees with an internal training program of quarterly conferences, webinars, and partner-run courses with certifications. The attitude matters more than the existence of a program though: strong entry-level talent today means strong senior talent tomorrow. If you want to better understand which tasks you should delegate to AI and which tasks to protect, it’s a conversation we have with clients all the time. If you want to think through it together, reach out to our advisory unit.