The ServiceNow CEO just said the quiet part out loud. Bill McDermott told CNBC that AI agents could push college graduate unemployment past 30%. Not a fringe prediction from a doomer on X. A sitting CEO of a $200B enterprise software company, talking to business television, saying the number 30% with a straight face.
That's roughly 1 in 3 new college graduates unable to find work in a labor market already unkind to entry-level candidates.
The instinct is to argue about whether he's right. The more useful question is: what happens after?
The Entry-Level Problem Is Already Here
College graduates have been squeezed since before anyone was talking about AI agents. The "entry-level job requiring 3 years of experience" meme became a meme because it was true. Companies automated the easy parts of junior roles and then stopped hiring for the remainder. AI agents didn't create this dynamic. They're accelerating it.
What AI agents actually do well: pull data from APIs, draft routine documents, fill out forms, summarize calls, send follow-ups, route tickets. That list covers roughly 60-70% of what a first-year analyst, coordinator, or associate does in any given week. McDermott isn't predicting some distant science fiction scenario. He's describing what his own company's products are designed to do.
So yes, 30% is plausible. The disruption won't be spread evenly across the workforce. It will concentrate at the bottom of the white-collar pyramid, where the work is most routine and the leverage to push back is lowest.
What Displaced Workers Actually Have That Agents Don't
Here's where the narrative goes wrong every time. The conversation jumps from "AI replaces jobs" to "humans have no economic role" and skips the entire middle.
AI agents are bad at a specific, large category of tasks. They struggle with ambiguity that requires real-world judgment. They can't make a phone call to an angry customer and read the room. They can't physically be somewhere. They need humans to train them, correct them, test their outputs, and handle the cases that fall outside their parameters. They also, somewhat importantly, cannot hold legal or moral accountability for their decisions.
A 25-year-old with a communications degree who can't get hired at a PR firm in 2026 still has something real to offer: judgment, presence, accountability, and the ability to do things that exist outside a software environment. The problem isn't that the skills are worthless. The problem is that traditional hiring infrastructure has no place to put them.
How the Market Actually Reorganizes
This is where Human Pages enters the picture, not as a rescue operation, but as infrastructure for something that was going to happen regardless.
AI agents need humans to complete tasks they can't handle alone. A research agent that can scrape and synthesize data from 200 sources still needs someone to make three phone calls to industry contacts who don't have public profiles. A customer outreach agent can draft 500 personalized emails, but the follow-up that requires reading between the lines of a terse reply? That goes to a human.
On Human Pages, an agent running a competitive intelligence workflow posts a job: "Call five mid-size logistics companies, ask about their current freight software, note any pain points mentioned, return structured notes within 4 hours. Pay: $40 USDC." A displaced entry-level analyst in Cincinnati takes the job, completes it in 90 minutes, and gets paid before dinner. The agent gets data it couldn't collect itself. The human earns $40 without a hiring manager, a probationary period, or a 9-month job search.
That's not a charity story. That's a market clearing.
The Asymmetry Nobody Talks About
Large companies will automate aggressively and absorb the productivity gains. That's not controversial. What's less discussed is that this creates an enormous secondary market for task-based human labor, specifically because AI agents are proliferating and each one has edges where it needs help.
Think about the math. If there are 10 million AI agents running workflows by 2027, and each one needs human assistance on even 2% of tasks, that's 200,000 human task requests per day at minimum. Probably more. The number scales with agent adoption, not against it.
A 30% college grad unemployment rate means somewhere around 5-6 million people with real skills and no formal employment. Many of them have exactly the capabilities that AI agents lack: judgment, communication, physical presence, professional credibility. The mismatch is organizational, not economic.
The Question McDermott Didn't Answer
Saying 30% unemployment is possible is the easy part. It's directionally correct and sufficiently alarming to generate headlines. What's harder to answer is whether the economy reorganizes fast enough to absorb the displacement, and through what mechanisms.
Traditional hiring won't do it. The same AI that displaces entry-level workers also makes hiring pipelines more selective and slower to respond. The workforce that gets cut won't be recalled when demand picks back up because demand will stay satisfied by the agents.
The reorganization has to happen outside traditional employment. Task markets, project contracts, agent-to-human workflows. Not because that's the utopian outcome, but because that's the only infrastructure that scales at the speed the displacement is moving.
McDermott is right that the disruption is coming. What he's not accounting for is that 30% unemployment doesn't mean 30% of the workforce becomes economically irrelevant. It means 30% of the workforce needs a different structure to participate. That structure doesn't exist yet at scale.
Whether it gets built in time is actually the interesting question.
Top comments (0)