ServiceNow's Bill McDermott went on CNBC last week and said graduate unemployment could "easily go into the mid-30s in the next couple of years." The Fed puts it at 5.7% today. Whether he's right about the number matters less than what he's drawing attention to. It's that almost nobody pushing back on it is addressing the real problem underneath.

The conversation about AI and jobs has been stuck in a binary loop for two years now. AI will destroy jobs. No, AI will create new ones. Both sides wave their studies around and nobody changes their mind. The actual risk is subtler than either camp admits, and it's already visible if you know where to look.

AI isn't coming for all jobs. It's coming for the bottom rungs of the career ladder — software development felt it first, and now it's spreading everywhere. When those rungs disappear, the whole structure starts to wobble.

The evidence is piling up

A Resume.org survey published this month found that one in five companies have already stopped hiring entry-level workers because of AI. Not planning to. Already done. Nearly half expect to have eliminated entry-level hiring altogether by 2027.

BSI, Korn Ferry, and others are landing in the same territory. Junior roles — admin, research, briefings, back office — are the ones disappearing fastest. Half of large companies have already made those cuts. Korn Ferry's 2026 Talent Acquisition Trends report found that more than half of leaders plan to add autonomous agents to their teams this year, and 37% plan to replace entry-level roles with AI outright.

These are people already making hiring decisions, not futurists working from theoretical models. The only real argument left is about speed.

The pipeline problem nobody's talking about

Most of the conversation treats entry-level roles as a cost line — cheap labour that AI can now do faster. But those jobs were also training grounds, and that's the part getting lost.

The graduate who spent six months doing research summaries was quietly learning how the organisation thinks — which questions matter, which data sources are trusted, what the senior people actually care about versus what they say they care about. The junior analyst cleaning data was developing the pattern recognition that would eventually make them a senior analyst who could spot when something didn't look right. The output was mundane. The education was not.

This is tacit knowledge — the kind you can't put in a manual or a training module. It accumulates through exposure to real work, real mistakes, and real feedback loops. It's how people develop judgement, and judgement is the one thing AI still can't replicate.

I saw this play out over twenty years at Vertical Leap. The people who became genuinely good — the ones I'd trust to run a client account or push back on a strategy that looked right on paper but felt wrong — were the ones who'd spent their first year or two doing the unglamorous work. Pulling reports, sitting in meetings they didn't fully understand, watching how senior people handled difficult conversations. You can't shortcut that. The rote work was the delivery mechanism for something far more valuable.

Korn Ferry's report flags this directly: a leadership pipeline crisis is forming as companies race to automate without considering where their future managers actually come from. Eliminate those roles today and you're scrambling to hire expensive outside managers in a few years — managers who won't understand the company they're supposed to lead.

And the Korn Ferry data on executive readiness makes it worse. Only 11% of talent acquisition leaders say their executives are well-prepared to lead through the AI transition. The people making decisions about cutting junior roles aren't equipped to think through the second-order consequences of those decisions.

The companies that complicate the narrative

The layoff patterns of early 2026 tell a more nuanced story than the "AI kills junior jobs" headline suggests.

Block cut more than 4,000 employees in February — roughly 40% of its workforce — with Jack Dorsey citing AI automation as the driver. That's a broad, indiscriminate cut across levels.

Atlassian slashed 1,600 roles in March to fund its AI pivot. But Atlassian explicitly protected its most junior workers. CEO Mike Cannon-Brookes said the company focused on retaining "graduates, and Atlassians with transferable skills" — suggesting the cuts fell more heavily on mid-level and senior employees whose existing workflows were harder to reshape around AI.

Atlassian's logic is worth paying attention to. They're betting that junior people, having grown up with AI tools, will adapt faster than expensive mid-career staff. It's a contrarian bet, and whether it pays off depends entirely on whether those junior hires have somewhere to learn the things that AI can't teach them.

The BSI survey found a striking admission buried in the data: 56% of bosses say they're lucky to have started their career before AI transformed their industry. And 43% acknowledge they wouldn't have developed the skills they have today if AI tools had been around when they started. They know the ladder matters. They're pulling it up anyway.

The taxation argument won't save us

Andrew Yang's recent proposal to shift taxes from labour to AI is getting traction, and the logic is sound in principle — if AI is displacing workers, the tax system shouldn't penalise companies for hiring humans. Even Dario Amodei, CEO of Anthropic, has said his company should be taxed on AI automation.

But taxation is a redistribution mechanism, not a training mechanism — and as I've argued, the training mechanisms themselves have a dismal track record. You can fund universal basic income, shore up social safety nets, and ease the financial pain of displacement. What you can't do with a tax is teach a twenty-three-year-old how to exercise judgement in ambiguous situations, how to read a room, how to recognise when the data is technically correct but the conclusion is wrong. Those skills require years of structured exposure to real work — the exact kind of work that's being automated.

The policy conversation matters, but it's downstream of a more fundamental question that every company cutting junior headcount needs to answer: where are your senior people going to come from in five years?

What companies should actually be doing

A handful of organisations are starting to think about this seriously. The approach that makes sense borrows from the apprenticeship model — not the Victorian kind, but a modern version that uses AI to accelerate learning rather than replace it.

Instead of giving a junior analyst a stack of reports to summarise manually, you pair them with an AI tool and focus their time on questioning the outputs, spotting what's missing, and understanding why the analysis matters. If the AI literacy gap taught us anything, it's that most people underuse these tools because nobody showed them what's possible. Junior hires don't need to repeat that mistake. The rote work goes away, but the learning accelerates — because they're engaging with higher-order tasks from day one instead of spending their first year on mechanical work that teaches patience more than skill.

This requires something most organisations aren't currently set up to provide: deliberate, structured mentorship that's designed around the reality that AI handles the mechanics. The junior person's job isn't to do the grunt work any more. It's to develop the discernment that the grunt work used to build — just through a different, faster path.

It also requires resisting the temptation to count those junior hires purely as a cost centre. The maths looks obvious when you're comparing a graduate's salary to the price of an AI agent subscription. But that comparison ignores the compound value of a person who understands your business deeply after five years — someone you don't have to recruit externally, onboard expensively, and hope figures out the culture before they make a costly mistake.

The companies that handle this well treat AI-augmented junior roles as an investment in the kind of human capability that AI makes more valuable, not less. The research on AI-augmented cross-functional teams is striking — they're significantly more likely to produce breakthrough ideas. But you don't get teams like that if you've spent five years not hiring the people who'd form them.

The real graduation cliff

McDermott's 30% number might be hyperbolic. Or it might be conservative — nobody really knows, which is part of the problem. But the graduate unemployment rate isn't the cliff that matters most.

The real cliff is quieter. It's the one where organisations look up in 2030 and discover they have a missing generation of mid-career talent. People who should have spent the late 2020s learning how the business works, developing the instinct for when something's off, accumulating the institutional knowledge that makes an organisation function. Instead, those people were never hired — or were hired into roles so hollowed out by automation that they never learned the things that matter.

You can't backfill institutional knowledge. You can't fast-track judgement. And you can't hire your way out of a talent gap you created by treating entry-level roles as a line item to be optimised rather than a pipeline to be maintained.

Before you cut another junior role, ask one question: who's going to replace the senior person who leaves in three years? If the answer is "we'll hire externally," congratulations. You've described the problem, not the solution.

Quick answers

Will AI eliminate entry-level jobs?

Not all of them, but the trajectory is clear. One in five companies have already stopped hiring for entry-level positions because of AI, and nearly half expect to eliminate those roles entirely by 2027. The jobs most at risk are those involving routine research, data processing, admin, and first-draft content creation.

What's the real risk of cutting junior roles?

The long-term risk isn't cost savings gone wrong. It's a leadership pipeline crisis. Entry-level roles are where people develop the judgement, institutional knowledge, and contextual understanding that makes them effective senior employees. Remove the training ground and you'll face a talent cliff in five to seven years that no amount of external hiring can fix.

How should companies use AI with junior employees instead of replacing them?

The smartest approach is pairing junior hires with AI tools so they engage with higher-order thinking from day one. Instead of spending a year on mechanical tasks, they learn to question outputs, spot gaps, and understand why analysis matters. The rote work disappears, but the learning accelerates — if the organisation provides structured mentorship around it.

What is Andrew Yang's AI tax proposal?

Yang proposes shifting taxes away from human labour and onto AI automation, and argues the tax system shouldn't penalise companies for hiring people. Even Anthropic's CEO supports being taxed on AI automation. It has merit as a redistribution mechanism, but it doesn't solve the training and skills development problem that entry-level job losses create.