New data backs parts of the Anthropic CEO’s prediction. But the industry’s response reveals a deeper problem.
Getty Images
In May 2025, Anthropic CEO Dario Amodei told Axios that AI could wipe out half of all entry-level white-collar jobs within five years. That prediction sparked a firestorm. When I spoke to experts about it, the pushback was swift. Andy Thurai, field CTO at Cisco, told me bluntly that “The AI providers — Anthropic, OpenAI, consultants — have to say extreme things to gain attention and instill FOMO.”
Eight months later, Amodei is back and he’s gone bigger, doubling down on his position. In late January, he published a 20,000-word essay titled “The Adolescence of Technology,” warning that AI systems smarter than Nobel laureates could arrive by 2027, that autonomous AI has already shown signs of deceiving its creators during testing and that AI-enabled dictatorships are a real possibility. He repeated the 50% jobs claim and called this moment humanity’s “rite of passage.”
But what’s changed between then and now isn’t exactly his warning. It’s the context around it.
The Data Moved
When Amodei first made his prediction last year, the evidence was thin. But that has since shifted. An MIT study published last November, using a simulation tool called the Iceberg Index, found that AI can already technically replace 11.7% of the U.S. labor market — roughly $1.2 trillion in wages across finance, healthcare and professional services. The visible disruption in tech layoffs accounts for just 2.2% of that exposure. The rest is building quietly beneath the surface, in HR, logistics and office administration.
The consulting firm Mercer reported that 40% of employees now fear losing their jobs to AI, up from 28% in 2024. IMF Managing Director Kristalina Georgieva said at Davos that AI is “hitting the labor market like a tsunami, and most countries and most businesses are not prepared for it.”
So, without a doubt, parts of Amodei’s thesis are landing and there seems to be an industry consensus about some of his concerns. But Amodei’s timeline remains the sticking point.
The Credibility Gap
Amodei predicted in early 2025 that AI would be writing 90% of code within six months. That turned out to be true inside Anthropic, as the company’s own Cowork tool was almost entirely written by Claude. But across the broader software industry, the figure was 25% to 40%. For non-tech companies, it was even less. As Fortune’s AI editor Jeremy Kahn wrote, Amodei “may have a skewed sense for how quickly non-tech companies are actually able to adopt technology.”
That pattern of seeing the future clearly from inside the lab but misjudging how fast it reaches everyone else matters. I think it’s the difference between a useful warning and a self-serving one. Anthropic is valued at $350 billion. And, as Kahn noted, the essay reads as “as much a novella-length marketing message as it is an impassioned prophecy.” The solutions Amodei proposes — constitutional AI, interpretability research and safety-first development — map directly onto Anthropic’s product roadmap.
At Davos, the skepticism was open. Scott Galloway, the NYU marketing professor and tech investor, argued that every previous technological innovation had created more jobs than it destroyed. Google DeepMind CEO Demis Hassabis put the probability of human-level AI within the decade at just 50% — far more cautious than Amodei’s one-to-two-year window. Deutsche Bank analysts warned that “AI redundancy washing will be a significant feature of 2026,” with companies blaming AI for cuts driven by other factors entirely.
An Industry Still Split
If Amodei’s warning is a fire alarm, the industry’s response looks less like an evacuation and more like a committee meeting about exit signs. The most immediate question his essay raises is simple: Whose job is it to act?
Josh Hart, co-founder and chief product and technology officer at YuLife, argued that regulation is the wrong starting point entirely. “We can’t simply regulate our way out of this transition,” he said. “The music industry didn’t beat piracy with lawsuits; they beat it with Spotify. We need to build an AI ecosystem so safe and convenient that the dangerous alternatives become obsolete.” In Hart’s view, the market will self-correct, if companies design better products.
Judah Taub, managing partner at Hetz Ventures, sees the problem through a different lens. “When even the CEOs of the biggest AI companies admit the future is ‘cloudy,’ and that they don’t really know where this is heading, that should worry us,” he said. Taub called for international rules comparable to nuclear nonproliferation agreements and tax structures that reward safer AI models — the kind of heavy coordination that markets alone have never produced.
Gil Perry, CEO of D-ID, landed somewhere in between. “The next phase of AI shouldn’t be defined by extreme hype or crippling fear, but by accountability,” he told me, arguing that trust needs to be built into products “not because of government regulations, but because of basic responsibility.” It’s a reasonable position. But it also depends entirely on companies policing themselves, which history suggests they tend to do right up until the moment it becomes expensive.
Eight months after Amodei first raised the alarm, the people closest to this technology still cannot agree on who should act, how, or when — and I think that’s the real story here.
The Response So Far
Governments have started to respond meaningfully, with states like Illinois, Texas and Colorado regulating how companies use AI in hiring decisions, focused on bias and transparency. But it addresses a different problem than the one Amodei is raising: What happens to the millions of workers whose roles don’t get filled by biased algorithms but simply stop existing altogether?
Companies like HCLTech have trained over 100,000 employees in generative AI, and the World Economic Forum estimates that 1.1 billion jobs will need to be transformed by technology in the next decade. But according to Gartner, 120 million workers globally are at risk of redundancy because they’re unlikely to receive the reskilling they actually need.
What To Watch
The test case may already be in front of us. ServiceNow CEO Bill McDermott vowed at Davos not to lay off employees, even as his 30,000-person company deploys agentic AI to automate job functions. It was one of the strongest public commitments any major CEO has made on AI and jobs. But every quarter that AI capabilities improve and competitors cut headcount, that promise gets harder to keep.
McDermott’s pledge is worth watching because it represents the exact tension Amodei’s essay describes: A leader who sees what’s coming and is trying to manage the transition responsibly, under pressure from a market that rewards speed over caution.
If a company of ServiceNow’s size eventually reverses course, it won’t just be one CEO breaking a promise. It will signal that the industry’s informal approach to managing AI displacement — goodwill, voluntary commitments and case-by-case decisions — isn’t holding.
Amodei has now sounded this alarm twice. The data behind it is stronger than it was eight months ago. And the industry’s answer is still the same: We’ll figure it out. That should concern us more than any single prediction about jobs.