The Vanishing First Rung of the Career Ladder
A leaked internal Anthropic employment report contains a number more striking than any benchmark of AI capability: the share of entry-level staff among new technical hires fell from 27% in early 2025 to just 9% in 2026.
The report was published on Anthropic's website on October 16, in two parts: one on the company's internal employment structure, and one comparing Anthropic employees' AI expectations with those of the general US public. Every figure deserves scrutiny, because each comes with caveats.
Group One: Working Hours
By March 2026, the total working hours of AI agents inside the company exceeded those of human employees. By August, the agents that employees personally direct were doing six times the work of their owners' estimated working hours. Note the qualification: this counts only employee-directed agents—adding company-orchestrated agents would push the figure higher.
Group Two: Headcount Shares
The report contains two distinct calculations:
- New hires: entry-level share of new technical hires fell from 27% to 9%.
- Existing workforce: the stock of junior employees in technical departments fell from 21% to 14%.
- Anthropic reports (anthropic.com/research/employment-at-anthropic and expectations-for-the-future, 2026-10-16): new-hire entry-level share 27%→9%; existing junior share 21%→14%; technical roles 61%→56%; one-third of ML managers self-report AI-driven hiring cuts; absolute technical headcount still rising; agent hours at six times (limited to employee-directed agents).
- Jack Clark's Per Jacobsson Lecture, IMF/World Bank annual meeting, Bangkok, October 17: "the singularity has begun" (multiple sources).
- Stanford "Canaries in the Coal Mine?" (Brynjolfsson et al.; August 2026 update widened the young-worker employment gap to 19%).
- IBM's tripling of US entry-level hiring announced 2026-02-12 (Bloomberg).
One governs the entrance, the other the stock—both moving in the same direction, with the entrance falling faster. There is also a bigger picture: technical roles dropped from 61% to 56% of the company, diluted by aggressive hiring in sales, finance, and recruiting.
The most important caveat: the company is still expanding, and absolute technical hiring is rising. What fell is the share, not the absolute number. In other words, they are still hiring—but fewer of those positions go to newcomers.
About one-third of machine learning managers and one-fifth of software engineering managers admit reducing hiring because of AI. The reason is blunt: tasks a newcomer once did—writing tests, checking documentation, fixing small bugs, running scripts—are now handled effortlessly by agents, with no onboarding required.
"The Singularity Has Begun"
The day after the report's release, Anthropic co-founder Jack Clark delivered the Per Jacobsson Lecture at the IMF/World Bank annual meeting in Bangkok. His theme in one sentence: the singularity has begun. His internal models estimate AI could push US GDP to the $44 trillion level. Within a single week, the same organization published data showing entry-level hiring cut to a third—and declared the singularity underway.
The Canary in the Coal Mine
Placed in the broader labor market, this echoes a famous Stanford paper titled "Canaries in the Coal Mine?": in AI-exposed occupations, employment decline appeared first among workers aged 22–25. The August 2026 update widened the gap from 13% to 19%—the canary is not only alive; the warning has escalated.
Another outlier deserves separate mention: IBM announced in February that it would triple US entry-level hiring in 2026, explicitly targeting roles AI cannot yet replace. While others pull the ladder away, IBM adds rungs proportionally—reclassifying entry-level roles from cost to investment.
The Deeper Question
This raises a question deeper than "AI steals jobs." The output of entry-level work was never the whole point. A newcomer's clumsy code or verbose documentation isn't valuable in itself—what's valuable is the process that turns a novice into an expert. Apprenticeship has survived for centuries not on lofty ideals, but because entry-level work combined two properties: cheap enough to hire for, and tedious enough that someone had to do it. AI has now removed the second property. The ladder remains; the rungs are disappearing.
Goldman-style investment bank CEOs have been asked the same question repeatedly in recent years: if AI handles all junior analysts' work, where will the partners of ten years from now come from? None has produced an answer. Anthropic's report simply advances the progress bar on that question: among its new hires, fewer than one in ten is a newcomer.