Every major technology shift in history has produced the same headline: the machines are coming for your job. The ATM was supposed to kill bank tellers. It didn't. Between 1980 and 2010, the number of bank teller jobs in the US actually grew, because cheaper transactions meant more bank branches, which meant more tellers. Josh Bersin just made a similar argument about AI, and he's right, though probably not for the reasons most people expect.
Bersin's thesis is straightforward: AI is a job-creation technology, not a job-destruction one. The historical pattern holds. When a tool makes something dramatically cheaper or faster, demand for that thing expands, new adjacent roles appear, and the total amount of work goes up. The question was never whether AI would create jobs. The question is what kinds, and for whom.
The Part Everyone Gets Wrong
The doom narrative assumes a fixed amount of work to be done in the world. AI takes some tasks. Fewer tasks, fewer jobs. Simple math, wrong model.
Work is not a fixed resource. A solo founder who can now ship a product with three AI agents instead of a six-person team doesn't just fire three people and call it a day. She ships twice as fast, enters two new markets, builds features that weren't economically viable before, and suddenly needs humans who can do things her agents cannot: negotiate with enterprise clients, handle regulatory filings in jurisdictions with unique requirements, do the kind of qualitative user research that requires sitting in someone's living room and watching them struggle with software.
Bersin puts the job creation number in the tens of millions globally over the next decade. That's not a random optimistic guess. It's based on historical adoption curves for general-purpose technologies, and AI is the most general-purpose technology we've ever built.
The jobs it creates are not the same jobs it displaces. That's the real problem, not the net total.
What AI Actually Needs Humans to Do
Here's a concrete example of what the new job creation actually looks like in practice.
An AI agent is managing customer support for a B2B software company. It handles 80% of tickets automatically. But the remaining 20% require a human: a client threatening to churn, a bug that needs someone to actually replicate it on a specific hardware setup, a billing dispute that requires reading between the lines of a contract written in 2019. The agent knows it can't handle these. So it posts the work to Human Pages, attaches the context, sets a budget, and a human picks it up within the hour.
That human is not doing a traditional support job. They're doing high-judgment, high-stakes work that the agent correctly identified as beyond its capability. They get paid in USDC, often within minutes of completing the task. No hiring manager, no resume screen, no two-week notice period. The agent needed a human, the human was available, the work got done.
This is the model Bersin is gesturing at when he talks about AI creating jobs. It's not that companies will hire more people in the traditional sense. It's that AI systems, operating at scale, will generate a continuous demand for human judgment that didn't exist as discrete, compensated work before.
The Numbers Are Not Small
Consider what it means when AI adoption reaches serious scale. The IMF estimated in 2024 that AI affects 40% of jobs globally. Goldman Sachs put 300 million jobs at risk of automation. These numbers get cited constantly as evidence of the apocalypse.
But both reports also note that most affected jobs are not eliminated, they're changed. A significant portion of the tasks within those jobs get automated, which frees the person to do more of the judgment work. That judgment work, when it overflows the capacity of the AI system, or when it's too sensitive to handle autonomously, becomes a discrete task that gets hired out.
If 300 million jobs are affected and even 10% of the judgment-intensive overflow becomes hired human work, that's 30 million new discrete work opportunities. Not jobs in the traditional sense. Something closer to a continuous market for human expertise and decision-making, clearing in real time.
That's not a utopia. It's also not an apocalypse. It's a labor market that looks fundamentally different from the one we built the 20th century around.
The Hard Part Nobody Wants to Say
Bersin is right that AI creates jobs. He's also diplomatically vague about who gets those jobs and whether they're good ones.
The people best positioned to benefit from the new AI-driven labor market are people who have skills that are legible to AI systems, available on demand, and capable of handling the specific gaps that agents can't fill. That requires a different relationship with work than most people currently have. It requires the ability to operate without a manager, to evaluate your own output, to price your own time.
This is not equally accessible. It favors people with existing expertise, digital fluency, and financial resilience. The person who spent 20 years doing a single repetitive job at a single company does not automatically get routed into the new market. That's a policy problem, an education problem, and a transition problem that optimistic job-creation forecasts tend to gloss over.
Acknowledging that AI creates work overall is not the same as saying the transition is painless or equitable. Both things are true at the same time.
What Actually Happens Next
The AI-hires-human market is not hypothetical. It's running right now, at small scale, and growing. Agents are posting tasks. Humans are completing them. The workflow is backwards from what the 20th century trained us to expect, but the underlying dynamic is the same: someone needs something done, someone else can do it, money changes hands.
The more interesting question is not whether AI creates jobs. It does. The interesting question is whether the humans who need those jobs can actually access them, and whether the work pays well enough and is consistent enough to build a life around.
We're early. The answer is not settled. But the direction is clear, and it's not the one the apocalypse forecast predicted.
Top comments (0)