Graduation Season 2025: The Year the Bottom Rung Disappeared
You don’t get to start at the beginning anymore
Every year, around this time, someone stands up in a robe and offers a few thousand 22-year-olds some words of wisdom and welcomes them into the “real world.” And for decades, the real world played its part: offer a handful of entry-level jobs, underpay generously, and teach just enough on the job to make people useful later. But now that contract is fraying.
The AI job apocalypse isn’t hypothetical anymore. It’s here.
Unemployment for recent grads is 5.8%. Which doesn’t sound terrible until you realize the overall national rate is much lower, and that this bump isn’t happening in, say, underwater basket weaving. According to Oxford Economics, the spike is concentrated in technical fields like finance and computer science. You know, the technical fields that used to be safe.
This gap signals a structural unraveling of how young people enter the workforce. Entry-level white-collar labor is no longer economically viable -because AI is cheaper, faster, and doesn’t need a mentor.
The Federal Reserve Bank of New York put it bluntly:
“The employment situation for recent grads has deteriorated noticeably.”
🧠 Entry-Level Work Is the Training Data
Anthropic, Google, and others are stress-testing agentic AI systems in Pokémon simulations. It sounds cute - but these are rehearsals for the real economy. These agents are learning how to complete multi-step objectives with no human in the loop.
Replace "gym battles" with "spreadsheet analysis," "code migration," or "client briefs," and you start to see the plan. Entry-level jobs aren’t just at risk - they’re the scaffolding AI is learning from.
🪓 It’s Not Just Job Loss. It’s Training Loss.
We keep asking: has AI replaced jobs?
That’s the wrong question. AI hasn’t just replaced jobs - it’s replaced the expectation that you’ll get trained. That someone will mentor you. That a job offer is also an education. AI advancements have removed the time and appetite for hand-holding.
So, no, it hasn’t eliminated employment. It’s just eliminated the bridge from potential to productivity. You’re either useful now - or you’re a lagging indicator.
🏢 AI-First Hiring Is Already Here - Just Not Always Out Loud
A few companies have said it out loud - Shopify, Klarna, IBM, Duolingo: test with AI before you hire a human. Many others haven’t said it, but they’ve made the shift.
“We’ve stopped hiring anything below L5.”
“One data scientist does what 75 used to.”
Even inside Anthropic, junior engineering roles are evaporating: “We tend more toward mid-level and above. Partially because junior work is changing.”
It’s not just about saving time. It’s about a new assumption: you start out fluent in AI, or you don’t start at all.
🔥 This Isn’t Just an Economic Issue - It’s a Safety One
We often treat AI safety as an abstract future problem: rogue agents, AGI alignment, existential risk. But as Kevin Roose pointed out: “A society with 20% youth unemployment isn’t a safe society.”
Mass job loss is an AI safety issue - one we’re not talking about enough. Sever the entry point to economic mobility, and you don’t just break careers. You destabilize the contract that keeps civil society intact.
🛠️ What Should Young Workers Do?
If the ladder’s missing its first rung, don’t wait for someone to fix it. Build a catapult.
In a world where entry-level jobs are disappearing, you’re left with 3 viable paths - all of which assume you’re self-directed, AI-fluent, and faster than the institutions trying to catch up:
Specialize in what AI still struggles with — ambiguity, persuasion, taste, human complexity.
Orchestrate AI workflows — become the person who directs agents, not the one replaced by them.
Leapfrog into solo projects or startups while the leverage window is open.
“If you can manage AI workflows, some companies will hire you straight into higher-level roles.” – Kevin Roose
This isn’t hypothetical. Anthropic’s founders expect the first unicorn with a single human employee to emerge as soon as next year.
Because agentic AI doesn’t just automate. It scales. Code, design, ops, sales - all can now be delegated to Claude-style agents. This isn’t software-assisted work. This is software doing the work. The implications for startups, venture capital, and labor economics are enormous.
But also the implications for young grads starting out are enormous.
To quote the sages of Twitter, you can just DO things.
AI didn’t kill jobs. It killed beginnings.