Vanishing entry-level jobs are the real impact of AI on the workforce

In May 2026, Sam Altman joined the Commonwealth Bank conference in Sydney and did something he doesn’t often do: he admitted he’d been wrong. The OpenAI founder and CEO had spent years warning about the imminent erasure of entire categories of white-collar work by generative AI. Now, he told the room, he was “delighted to be wrong” because the collapse he’d predicted hadn’t arrived. 

Altman was not the only leader to raise the alarm. In May 2025, Anthropic CEO Dario Amodei warned that AI could wipe out up to half of all entry-level white-collar jobs within five years and push unemployment to between 10% and 20%. 

Considering the Yale Budget Lab, which has tracked US labour data since ChatGPT launched, found no meaningful change in the occupational mix or unemployment that it could attribute to AI, a year later, the picture looked different. By the time Altman spoke in Sydney, both he and Amodei had softened the starkest versions of their forecasts. 

Altman deserves some credit for revisiting a prediction that didn’t play out as expected. In a field that tends to move on quickly, he chose to publicly examine where his intuition had fallen short. 

The pushback is that, while Altman asked the right question, he looked for the answer in the wrong place. The impact of AI on work was never going to announce itself in the national unemployment rate. The earliest signals show up in the day-to-day operational choices companies make, before they reach the economic data. 

Unemployment is the last place where the change appears 

What unemployment data can’t tell you is how many jobs were never created. The decisive moment comes earlier, when a company chooses not to add a role at all. 

That decision is where AI is already reshaping the workforce. Stanford’s Digital Economy Lab reviewed payroll records for millions of US workers and found that early-career workers aged 22 to 25 in the most AI-exposed jobs experienced a 19% relative decline in employment after late 2022. Their more experienced colleagues in the same occupations held steady or kept growing. 

This shows up in the wider graduate market too. Employment among software developers aged 22 to 25 fell by nearly 20% from its recent peak. 

According to the Federal Reserve Bank of New York, unemployment among recent college graduates reached 5.6% in March 2026, above the 4.2% rate for all workers. That reverses a pattern in which a degree gave young workers an edge over the wider workforce. 

AI isn’t eliminating existing roles for experienced white-collar workers, as Altman and Amodei predicted. Instead, it’s erasing the positions that once gave graduates their first foothold on the corporate ladder and let today’s experienced workers grow over time. The eventual implications of fewer entry-level roles, and of the experience creep working its way up the ladder are yet to be felt. 

I see this pattern in enterprise operations as companies rarely fire their support team and replace it with software, because doing so is disruptive, expensive and bad for morale. 

Instead, it keeps the team, hands the repetitive, high-volume work to AI, and then, when demand grows the following year, simply declines to open the fifteen new roles it would have advertised. Nothing shows up in the redundancy figures because no one has been let go, yet the cost of serving each customer keeps falling. 

The replacement story misreads what is happening 

People often talk about AI adoption and job losses as if one automatically produces the other. In practice, the relationship is shaped by how each company chooses to use the technology. 

The Stanford research points directly to this mechanism: the declines cluster in jobs where AI automates a task outright rather than assisting the person doing it. The pattern is not work vanishing but work changing shape. As AI takes on routine tasks, people are freed to focus on the decisions, exceptions and relationships where they add the most value. 

Voice shows the shift clearly 

Communications is one place where this is easy to observe. Voice carries a company’s hardest conversations: the billing dispute, the urgent appointment, the caller who wants to reach a real person. For years, it resisted automation because the older technology sounded robotic and tended to frustrate callers further. 

That barrier has fallen as modern systems hold a natural conversation, understand what a caller wants and carry the task through to completion instead of dumping them into a queue. When a company automates its routine calls, the underlying demand does not disappear; it is handled differently. 

The people who used to answer those calls take on the conversations that genuinely need a human. A new role appears beside them: the person who supervises, trains and improves the AI systems. AI removes the need for some roles while creating others. 

A recent Goldman Sachs analysis found that over the past year, AI reduced US monthly payroll growth by roughly 16,000 jobs. The losses fell on roles where technology substitutes for people, such as data entry and insurance claims work. 

In occupations where AI augments workers rather than replacing them, it added about 9,000 jobs a month. The reduction shows up as unfilled roles rather than experienced staff being let go, while demand for labour rises in new areas such as data centres and AI development. 

The rise of the forward-deployed engineer is one of the clearest signs of how AI is reshaping technical careers. The role barely existed a few years ago and is now becoming essential. These engineers do not simply write code; they work alongside customers, turning AI from an impressive demonstration into a dependable business tool, which demands a different skill set. 

Rather than decades of experience in a single niche, the advantage comes from quickly understanding workflows, solving problems in context and iterating against real feedback. As enterprise AI spreads across industries, this model will only become more common. I expect customer-side deployment and optimisation to shift from a specialist function into a defining feature of the engineering profession. 

So while AI is closing the door on many entry-level roles, it is not delivering the wholesale headcount collapse that Altman and Amodei once forecast. That should change how leaders think, because AI tools, like the people they work alongside, cost money to run, which raises the question of where they actually create value. Companies that treat automated interactions purely as a way to cut costs may be missing the larger return. 

Looking ahead, a tipping point seems likely by the end of 2026, with more than half of large contact centres deploying AI tools capable of handling complete interactions without human involvement. The companies that get the most from these systems will treat those interactions as a source of intelligence rather than a cost to contain. Every conversation tells them something concrete about what customers actually want, and that knowledge will start to shape products and decisions well beyond the call centre. 

What leaders should actually measure 

Altman made a second point in Sydney that drew far less coverage than his apology. He said the economic adoption of AI is still in its very early stages, and that leaders are right to keep asking whether it delivers measurable results. That is fair, and the answer is not in the national jobs report. 

If you want to understand AI’s impact, start with your own metrics: cost per task, hours per outcome, and headcount growth that never materialises. If those numbers improve while output grows, AI is already changing your workforce, whatever the macroeconomic data says. 

The change is already visible to any leader willing to read their own operations. By the time the national figures settle the debate, the companies that read theirs early will already be ahead.