Jason Furman once ran the White House Council of Economic Advisers. If he got the job back today, he says he would hand much of the grunt work to software.
In an interview with Axios on October 23, 2026, the Harvard economist said he would rely on AI tools over staff economists for a large share of the council’s day-to-day analytical work.
What Furman actually said
Furman framed the choice in stark terms. Given the option between current AI systems and a staff of 20 PhD or pre-doctoral economists for routine analysis, he would take the machines.
“I would somewhat regretfully and sadly choose the AI systems.”
His reasoning centers on speed and reliability. According to Furman, AI tools have become faster and more dependable for factual research, data synthesis and routine analysis, as long as they are properly supervised by fact-checking agents.
Furman also drew clear limits. He said human economists remain necessary for work that depends on interpersonal engagement, such as meetings and dealing with the press. He also acknowledged that AI lacks the creativity of the best human researchers. His argument is narrower than the online reaction suggested: the routine tasks, he says, can be done well by machines, while original thinking still belongs to people.
The website that backs up the claim
Furman did not just talk about it. On the same day as the interview, he launched Econvitals, a public website that pulls together US economic data in one place.
The site goes beyond a data dashboard. It offers analysis of whether individual Federal Reserve officials lean dovish or hawkish, which is economist shorthand for whether a policymaker tends to favor lower interest rates to support growth or higher rates to fight inflation.
Econvitals also publishes email-style research summaries and AI-generated economic papers. Furman said the project was built using tools including Anthropic’s Claude and OpenAI’s Codex.
Why Econ Twitter took notice
The comments landed hard in the loose community of economists who still gather on social media, the group once known as Econ Twitter. Discussion of Furman’s remarks spread quickly through policy and economics circles on October 23.
The debate split along a few predictable lines. Some focused on what the remarks mean for the profession. If routine analysis is automated, what exactly is a junior economist for? Others zeroed in on the idea of “value-add” skills. If a model can pull data and write a passable summary, the human premium shifts toward judgment, framing and the ability to ask the right question in the first place. A third thread looked further ahead, at economics training programs. Graduate schools and pre-doctoral pipelines are built around teaching students to do exactly the kind of work Furman says AI now handles.
The critique carries extra weight because of who is delivering it. Furman is not a Silicon Valley founder pitching a product deck. He chaired the CEA and currently co-teaches Economics 10, Harvard’s introductory economics course. He has also been active in discussions about technology’s effect on the economy since at least 2018.
What this means for policy work and the people who do it
The most immediate implication is for government and institutional research shops. The CEA is a small, elite operation. If its former chairman believes 20 economists could be largely replaced for routine tasks, other agencies, think tanks and central bank research teams face the same question.
The supervision piece is the real stress point. Furman’s endorsement depends on fact-checking agents catching errors. In a policy setting, a confidently wrong number can travel from a memo to a presidential briefing to a press conference.
For the AI companies themselves, the episode is a useful case study. Claude and Codex are being used here not as novelty chatbots but as the engine behind a public research product from a credentialed economist.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.