Ask whether the AI layoffs of the past two years were a catastrophic overreaction or a preview of a permanently smaller workforce, and the honest answer is uncomfortable. Both are true at once. Experienced workers are being rehired, often at a premium, because companies discovered that AI could not do their jobs alone.
Meanwhile, an entire generation of graduates is discovering that the jobs they were trained for were never advertised in the first place. This is the paradox defining the labour market in 2026, and it splits neatly into three acts.
The Boomerang
Through 2024 and 2025, corporate leaders stood on stage and told investors they were becoming “leaner” thanks to artificial intelligence (AI).
Behind the scenes, a different story was unfolding. Research from Forrester found that 55% of employers who made AI-driven layoffs now regret the decision, and separate surveys from Robert Half and Careerminds put the share of companies already rehiring for those exact roles at roughly 29% to 32%.
Gartner goes further, forecasting that half of all companies that cut customer service or operations staff in AI’s name will be forced to restaff those functions, often under new job titles, within the next year or two.
This is what analysts are calling the AI boomerang, where a role gets automated away, the automation falls short, and the company scrambles to rehire, usually at a cost.
Careerminds found that nearly a third of employers who rehired ended up spending more than they originally saved. Returning staff are also commanding higher wages, with some reports pointing to pay bumps of 20% to 35% above the original salary, because the new version of the job requires someone who can both do the work and supervise the AI doing it.
The reason is simple, even if executives were slow to grasp it. Generating output is not the same as exercising judgement. An AI model can draft a customer response or write a block of code in seconds.
It cannot always be trusted to decide whether that response is appropriate, or whether that code is safe to ship. Someone still has to review, route and approve the work, and that someone turned out to be expensive to lose.
Klarna is the case study everyone points to. The Swedish fintech built its AI assistant on OpenAI’s technology, claimed it could do the work of roughly 700 human agents, and cut close to 1,000 customer service jobs.
Within months, customer satisfaction had dropped sharply. Chief executive Sebastian Siemiatkowski later admitted the strategy had been too heavily skewed toward cost-cutting, and began rehiring human staff so customers could always reach a person if they wanted one.
Klarna only reached break-even in May 2026, a reminder that the savings from automation are not always as clean as the announcement suggested. Ford and Commonwealth Bank of Australia tell similar stories.
The quality problems and swelling call volumes forced both to bring experienced staff back, this time to train new hires and to fix what the AI could not.
None of this means generative AI is a failure. It means companies confused a tool for a workforce, and are now paying to correct the mistake.
The Silent Collapse
While experienced employees are being rehired, an entirely different and much quieter crisis is unfolding one rung down the ladder. Graduates are not being laid off. They are simply not being hired at all.
The numbers are stark. Entry-level hiring at the biggest tech employers fell by around 25% between 2023 and 2024.
A Stanford HAI study found that employment among software developers aged 22 to 25 has dropped almost 20% since its late-2022 peak, even as employment for developers over 30 kept growing.
Big Tech’s own data shows new graduates now make up around 7% of hires, down from roughly 32% before the pandemic. It is not that companies are firing juniors. They have simply stopped opening the door.
The mechanism is straightforward once you see it. Entry-level jobs have always been built around repetitive, well-defined tasks such as drafting boilerplate code, running first-pass research, handling routine queries.
These are precisely the tasks that large language models and AI agents now do fastest and cheapest. When a company decides it needs fewer hands on routine work, the junior role is usually the first one to disappear from the org chart, because it was built almost entirely out of the tasks AI now absorbs.
The human cost of this shift is easy to underestimate from a spreadsheet. A generation of students took on years of study, and often meaningful debt, in response to a decade of “learn to code” messaging from the very industry that is now telling them the entry point no longer exists.
Computer science graduates currently face unemployment rates close to double the national average. The traditional on-ramp into a corporate career, the place where a 22-year-old was once allowed to be slow, make mistakes and learn on the job, is narrowing fast.
The shift is also changing what employers increasingly mean when they advertise an “entry-level” role. Historically, those positions assumed candidates would arrive with theoretical knowledge and acquire practical experience over time.
Today, many employers expect applicants to already know how to work alongside AI systems, evaluate machine-generated output and navigate complex workflows with minimal supervision.
In effect, the definition of junior has shifted upward. Graduates are now competing for jobs that demand skills once associated with employees who already had a few years of experience.
This creates a difficult feedback loop. Companies argue they cannot find graduates with the right capabilities, while graduates struggle to gain those capabilities because the jobs where they would traditionally learn them have disappeared.
Universities can teach programming languages, marketing principles or financial analysis, but they cannot fully replicate the judgement that comes from solving real business problems, making mistakes and learning under experienced colleagues. Apprenticeship, whether formal or informal, has always been one of the hidden engines of economic productivity.
There is another consequence that receives far less attention. Innovation itself often depends on newcomers. Junior employees ask naive questions, challenge established assumptions and introduce ideas that veterans may overlook precisely because they have fewer preconceptions.
Organisations that become top-heavy risk losing not only a future leadership pipeline but also a valuable source of experimentation and fresh thinking. History shows that many breakthrough products and technologies emerged from teams that mixed experienced judgement with youthful curiosity.
The challenge, then, is not simply preserving jobs for the sake of employment statistics. It is preserving the ecosystem through which knowledge, responsibility and expertise are passed from one generation of workers to the next.
Without that transfer, businesses may discover that the skills they assumed could be bought later were never available to purchase at all.
The Coming Talent Drought
Bring the two threads together and a harder question appears. If companies automate away their junior employees today, where do their senior employees come from in five or ten years?
Experience is not something a model can generate. It has to be lived, mistake by mistake, on the job. Cut off that pipeline for long enough and there will be nobody left with the judgement that companies are currently paying a premium to rehire in Act One.
Some companies have already worked this out. IKEA’s parent company, Ingka Group, deployed an AI assistant called Billie that now handles 47% of customer inquiries.
Rather than cutting the 8,500 call-centre workers whose routine workload had disappeared, Ingka retrained them as remote interior design consultants. That new advisory channel now generates around 1.3 billion euro a year, turning what had been a cost centre into one of the company’s fastest-growing revenue lines.
IBM has taken a similar view. Its chief human resources officer, Nickle LaMoreaux, announced the company would triple its entry-level hiring in 2026, including for software development roles that AI is supposedly capable of doing.
IBM’s response was not to ignore automation but to rewrite what junior roles actually involve, shifting new hires toward client interaction, systems thinking and reviewing AI output rather than producing routine code from scratch.
The lesson from both examples is the same. The companies most likely to come through this transition intact are not the ones using AI to shrink their headcount.
They are the ones using AI to free up human capacity for the judgement, taste and accountability that no model can yet supply, and then investing that freed capacity in the next generation of talent rather than cutting it loose.
The AI boomerang shows what happens when companies mistake automation for elimination. The junior hiring freeze shows what happens when nobody notices the ladder has lost its bottom rung.
Left unaddressed, the two problems will meet in the middle of the next decade, when the experienced workers being rehired today retire, and there is no one left who was ever given the chance to replace them.