Daron Acemoglu, who won the Nobel Prize in economics in 2024, says three developments are shaping how AI actually hits the labor market: the rise of agentic AI, the hiring of in-house economists by frontier labs, and the absence of widely usable AI applications. He laid out the framework in an interview published May 11, two years after his paper arguing that AI would deliver only a small boost to US productivity rather than the white-collar overhaul Big Tech CEOs had promised.
Acemoglu's measured view has lost ground in public discourse since 2024. A California gubernatorial candidate said last week he wants to tax corporate AI use and pay victims of AI-driven layoffs, and Senator Bernie Sanders has folded the AI jobs apocalypse into his rallies. The data still cuts the other way: studies have repeatedly found no measurable effect of AI on aggregate employment rates or layoffs since Acemoglu began publishing on the question.
On agents, Acemoglu is blunt. "I think that's just a losing proposition," he says of pitching agentic AI as a one-to-many replacement for human workers. He frames agents as tools that augment specific pieces of a job rather than substitutes for the job itself.
Key facts
- 01Daron Acemoglu won the Nobel Prize in economics in 2024 after publishing a paper estimating AI would give only a small boost to US productivity.
- 02OpenAI hired Duke's Ronnie Chatterji as chief economist in 2024 and paired him with Harvard's Jason Furman, a former Obama advisor.
- 03Anthropic has convened 10 leading economists; Google DeepMind hired University of Chicago's Alex Imas as director of AGI economics last week.
- 04Acemoglu has researched AI since 2018 and points to x-ray technicians juggling 30 tasks as a test case for agentic AI.
- 05A California gubernatorial candidate proposed last week taxing corporate AI use to pay victims of AI-driven layoffs.
His example is an x-ray technician, who juggles 30 different tasks ranging from taking patient histories to organizing mammogram archives. A human switches between formats, databases, and working styles without thinking; an agent would need a distinct tool or protocol for each handoff. Acemoglu has been researching this task-decomposition question since 2018, and his bet is that many jobs survive an AI takeover precisely because agents cannot yet orchestrate between tasks the way humans do.
“An x-ray technician juggles 30 different tasks, Acemoglu notes — and whether agents replace workers depends on whether they can orchestrate between tasks the way humans do naturally.”— Jaeden Schafer
The frontier labs are racing in the other direction, competing publicly on how long their agents can run autonomously without errors. Acemoglu's point is not that the benchmarks are fake, but that they measure the wrong thing relative to how jobs are actually structured.
The second shift is a hiring spree that has nothing to do with researchers. OpenAI brought on Ronnie Chatterji from Duke University as chief economist in 2024 and announced last year that Chatterji would work with Harvard's Jason Furman, formerly an advisor to Barack Obama, on AI and jobs research. Anthropic has convened 10 leading economists for similar work. Google DeepMind announced last week that it hired Alex Imas from the University of Chicago as its director of AGI economics.
"It makes sense," Acemoglu says of the trend, noting that AI companies are aware public skepticism about AI is growing, largely on jobs grounds, and have direct incentives to shape the economic narrative around their products. He cites OpenAI's recent industrial-policy proposal as one example of how labs are entering economic-policy debates directly.
His concern is conflict of interest. "What I hope we won't get," Acemoglu says, "is that they're interested in economists just to further their viewpoints or further the hype." The emerging field of AI economics may end up dominated by research funded by the companies with the most to gain from favorable conclusions.
The third signal Acemoglu is watching is the slow build-out of practical AI applications. He draws a contrast with the software that powered earlier tech transformations. "Anybody could install these on their computer and get them to do the things that they want them to do," he says of PowerPoint and Word, and they spread accordingly. "We have not seen the development of apps based on AI that have the same usability," he adds.
Chatbots are easy to open but hard to extract durable productivity from, which Acemoglu argues is part of why AI has not yet moved the macro needle on jobs or output. The signal he is watching is the arrival of AI-native applications that average workers can pick up and use without retraining — the equivalent of installing Word, not the equivalent of learning to prompt.
Acemoglu concedes the picture is messy. "There's a huge amount of uncertainty," he says, and the next several years will produce conflicting signals: anecdotes of college graduates struggling alongside flat aggregate productivity numbers. Some previously skeptical economists have shifted toward the apocalypse camp; the data has not yet followed them.
The labs are betting the opposite way, and the economist hires are how they intend to make that case. If Acemoglu is right that agents stumble on task orchestration and that AI's killer app has not yet shipped, the frontier labs are buying economic narrative ahead of economic reality — which is a defensible commercial move, but one worth labeling as such. The interesting tell over the next year will be whether the in-house economists at OpenAI, Anthropic, and DeepMind publish findings that diverge from their employers' product roadmaps, or whether they don't.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
