Washington Post Intelligence just published a conversation with Stanford economist Erik Brynjolfsson on why an AI “job apocalypse” is unlikely. I work with Erik through my work with the AI Fund, and I completely agree with where he is coming from.
There is too much precedent from past waves of automation to treat every new capability as the end of work. Some jobs will go away. Others will change shape. Over time, the pattern we keep seeing is a net gain for people who get in front of the trend, learn the tools, and help redesign how the work gets done.
What the evidence actually shows
Speaking with WPI Deputy Editor Yun-Hee Kim on the sidelines of Ai4 in Las Vegas, Brynjolfsson pushed back on the loudest forecasts. Tech CEOs from Anthropic’s Dario Amodei to Elon Musk have warned that AI could lead to massive job losses. Erik’s reading is calmer and more useful. He does not see widespread job displacement associated with AI. He does predict reduced demand for entry-level workers where the technology can substitute for routine knowledge work, while demand for experienced workers such as senior coders stays high.
With colleagues at Stanford’s Digital Economy Lab, he updated Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. They still do not see widespread, economy-wide displacement. Updated ADP payroll data do show a real entry-level squeeze: employment among workers ages 22 to 25 in highly AI-exposed occupations now stands about 19 percent below where it would be if it had kept pace with similarly aged workers in less-exposed occupations. That is a signal about the career on-ramp. It is a long way from “work is over.”
Three updates since the August 2025 Canaries report are especially clear in the interview. First, this is a hiring story more than a firing story. Companies are quietly not opening junior requisitions. Second, how you use AI matters more than whether you use it. Third, roughly 96 percent of firms using AI report no major change in headcount. On the upside, nonfarm business productivity growth is running over 2 percent, the best sustained stretch since the late-1990s boom. The productivity J-curve Erik has argued about for a decade is starting to show up in the data.
Complement, do not only cut
“Most senior executives have an overemphasis on replacing humans in cost-cutting,” Brynjolfsson says. “There’s nothing wrong with saving money but, the reality is, in most cases, they’re actually better off looking for opportunities,” with AI. That line should be on more whiteboards than the apocalypse slides.
He organized a letter to policymakers last month with a simple three-part message: the technology is getting more powerful, the economic stakes include both higher living standards and real disruption, and we have to act now. The constructive move is institutions, policy, and research that steer AI toward complementing people so it creates more jobs. A common misconception among managers, he notes, is that AI has to reduce jobs to be effective. You can raise employment and productivity together.
Tax policy that heavily favors capital over labor quietly guides firms toward replacement. The better economic instinct is extension: have people do new things they could not do before. Shared prosperity is a design choice, not an accident.
Keep the on-ramp open
If companies stop hiring at the base of the pyramid, they will not have experienced people later when they need them. Law firms and investment banks cutting junior intake may look rational in the short run. In a few years, the missing mid-career bench becomes the problem. One large technology company Erik mentioned is using AI as tutors to get new hires up to speed faster. That is the right spirit: keep the pipeline, compress the learning curve.
Through Workhelix, which he co-founded, he also presses leaders to stop flying blind on ROI. Seats, licenses, and chats per week tell you almost nothing. Measure tasks, not whole job titles, because AI almost never takes an entire job bundle. Score which tasks are amenable, where the value sits, how much time is saved, and whether quality holds. Customers such as Nasdaq, Autodesk, and LogicMonitor use that lens to decide where to deploy.
His 2030 sketch fits the same calm frame. Models will be enormously more capable. Productivity should run faster. Unemployment can still look historically unremarkable, because the binding constraint has always been how fast organizations rebuild processes, retrain people, and rewrite work. The worry he names, and I share, is a labor market that keeps overall employment while quietly closing the on-ramp for people starting their careers. That is exactly why we have to get in front of the trend now.
Morale is an economic input
People like Erik are pointing the way with data, not vibes. Doomsayers, and those who put out negative headlines mainly to gain engagement, ultimately damage morale and the economy as a whole. Fear freezes hiring, freezes learning, and freezes the experiments that create the next roles. A clear-eyed reading of Canaries, the productivity numbers, and the hiring-not-firing pattern is more useful than another apocalypse thread.
Invest in self-education and in understanding the landscape, and there is much less to fear. Learn how the models fail. Learn how seniors in your field use them as complements. If you lead a team, design for augmentation, keep humans where responsibility sits, and measure whether AI is creating capacity or only cutting cost.
Time to dig deeper
Start with the WPI Conversation with Erik Brynjolfsson. Then read the updated Canaries facts from Stanford’s Digital Economy Lab. Ask yourself where AI is already substituting for routine knowledge work in your field, where experienced judgment is still scarce, and whether your organization is still hiring and teaching at the base. Write that down. Share it with a junior colleague. The future of work gets less frightening when we treat it as a curriculum instead of a cliff.