AI can do the tasks. The mission still needs people, especially young ones.
AI is coming for the economics profession itself. The debate lands differently when you hear your job, or the job that started your career, described as replaceable by AI.
That was my experience a few weeks ago when I read an interview with Jason Furman. He argued that current AI models could replace most of the economic staff he had at the Council of Economic Advisers. I worked as a senior economist at CEA, on detail from the Fed, for Furman in 2015-16, so that got my attention. His suggestion that CEA’s junior staff, in particular, should be replaced was even more troubling. Juniors at CEA, like research assistants or pre-docs elsewhere, are the profession’s pipeline. My research assistant job at Brookings was my first step to becoming an economist.
If AI can do it, why do we need a person?
I will be honest: my first reaction to Furman was emotional. CEA was the most consistently overworked I’d ever been, so the idea of AI replacing my role was both annoying and appealing. I also have a bit of a mama bear instinct for the next generation of economists. Furman later admitted he had been overly provocative in calling for AI to replace CEA’s 20 or so economists; his follow-up note suggests a skeleton crew of roughly half that, with most of the juniors gone. Furman’s follow-up note is well worth reading, and the questions he is posing are not an idle thought experiment.
Furman recently embarked on an experiment to build AI systems that could produce research papers, blogs, briefings, and data dashboards on command. (You can find many of these tools on his new Econvitals website.) As a former CEA Chair, Furman has a good sense of the quality and timeliness CEA work requires, and in his view, the AI outputs are good enough to do much of CEA’s work:
Some of the AI-generated papers, blogs and briefings are cutting-edge. Some need significant work and have big deficiencies that need to be remedied by humans. And some of the longer papers might even be garbage. But all of them were done by AI in 2026 (it will get better), by systems I architected (I’m decent at this but many people are much better), and with basically no human input or iteration beyond architecting the system and then entering a simple prompt.
It’s a natural next question: if AI tools can do the work, why keep employees who used to do it? Here is Furman’s calculus on the junior staff at CEA, who are either recent college grads or graduate students, even giving some consideration to the pipeline:
But we also need to be honest about the tradeoffs. CEA was paid for by taxpayers. I could maybe justify hiring someone at $75,000 who produced $50,000 a year of output on the grounds that it would help train and mentor people to contribute more in the future. But I could not justify jobs where the person made only small contributions here and there and the bulk of what they did could be done by an AI, just because it would be good for them and the world more broadly if they got that training. Moreover, it is not clear what it means to get “training” if all you end up doing is botsitting.
Agree or disagree, Furman is raising questions that many professions will need to grapple with. To be transparent, I use AI models for my data prep and coding and do not have a research assistant. The AI is doing work I used to do myself, or that simply hadn’t gotten done. Budgeting $2,400 a year for AI in my consulting work was possible; adding an employee, even a junior one, was not. The AI did not replace a junior. It replaced nobody.
Can AI really do what matters?
Furman is using the task lens that is now ubiquitous in AI research. Take a job, break it down into its tasks, and evaluate how many of them AI can do. It is a useful starting point, but jobs are more than tasks. A task list tells you what a person does on a given day. It does not tell you why the organization hired them. Jobs also serve a mission.
In my Bloomberg Opinion piece last week, I examined how AI stacks up against CEA’s mission:
Created in 1946, the CEA’s mission is to offer the president “objective economic advice on the formulation of both domestic and international economic policy.” The most important version of that: Stop bad ideas. CEA has a storied history as the wet blanket of the White House. Kenneth Arrow, a Nobel laureate, often recounted that the most valuable thing he did in his life was a memo he wrote as a CEA staffer in the 1960s opposing a plan to widen the Panama Canal with an atomic bomb.
When I was at CEA, Arrow visited, and I vividly remember him telling this story. A Nobel laureate, he chose to tell a story about a memo. He made a point of telling it to CEA Chairs. The memo isn’t the point; what’s in it is. Would AI have made Arrow’s argument? An AI model asked to evaluate the plan would have evaluated the plan. Arrow decided it needed to be stopped.
AI models tend not to call out bad ideas; people do. Not everyone, of course. Dissent inside the White House tends to come from a distance. The chair and members are political appointees, chosen in part because they share the president’s priorities; the staff economists, many of them on leave from universities or detailed from the Federal Reserve or cabinet agencies, are not. They have less to lose by saying no, and in my experience they say it more often.
The same goes for distance from the profession. In my years working with junior staff at the Fed, CEA and elsewhere, the people who had not yet absorbed every tenet of economics were often the best at noticing when something did not add up. It is rarely as dramatic as Arrow’s memo. More often it is a confused look that tells you a project needs rethinking, or an argument needs refining.
First, it is straightforward to see how stopping a bad policy idea (or improving a good one) could pay CEA’s entire staff’s salaries many times over. Second, it’s not clear that you can serve that mission well with only senior staff. Furman argued that the CEA Chair would still be necessary since AI can’t go on television or talk in meetings, but the pushback from CEA comes from all levels, and frankly, it would be overwhelming to leave that role to a few people at the top. The juniors may have less experience to draw on, but they are often very perceptive about what needs improving. I saw a junior write an innovative policy speech for Furman, a task that at the Fed would have gone to a seasoned economist. But by the time you’re seasoned, you may no longer be pushing the innovative ideas. I have always learned a lot from my junior colleagues. If CEA had a smaller staff due to AI, I would strongly push back on cuts that fall disproportionately on juniors.
My Bloomberg piece also touched on creativity. Sometimes the president asks questions we don’t have a ready answer for. I had a project like that in 2016, when The Atlantic featured the $400 question from the Fed’s Survey of Household Economics and Decisionmaking (SHED) in “The Secret Shame of Middle-Class Americans.” AI would have helped us crunch the other data faster, but the basic question of why so few people have a financial cushion didn’t have an answer anywhere for an AI model to find then. The back-and-forth with memos and the president’s follow-up questions was intense. It stuck with me. Soon after I went back to the Fed, I took a role overseeing the SHED. Since then, I’ve spent many more hours on the $400 question, collecting new data and reviewing related research. Questions you just can’t quite answer have a way of pushing you to do more.
Staff using AI would have more time for the new questions. That would require leadership to rethink tasks and responsibilities. The goal would not be ten times the memos; it would be better memos. Even there, quality has limits. The time might be better spent communicating the analysis: videos, graphics, other formats. White House comms often referred to CEA publications as “vegetables.” Finding ways to get people to enjoy vegetables would be a real value-add, and there was little time for it before. And the junior staff would likely have the best ideas on how to reach a wider audience, and certainly the energy to experiment.
In closing.
The Council of Economic Advisers is one corner of the economics profession. Debates about how to use our scarce resources in doing economic analysis come up in every setting: the Federal Reserve, think tanks, academia, and the private sector. Paul Goldsmith-Pinkham had a thorough post on the future RAs and TAs in an AI world. He makes a compelling case for human judgment when using AI tools. Markus Brunnermeier has hosted several conversations about how AI can reshape economic research, and a recent dust-up about using AI to score points in economic arguments shows economists have a long road ahead in establishing norms on AI use.
The Council of Economic Advisers is one corner of the economics profession. Debates about how to use our scarce resources for economic analysis come up in every setting: the Federal Reserve, think tanks, academia, and the private sector. Paul Goldsmith-Pinkham had a thorough post on the future of RAs and TAs in an AI world. He makes a compelling case for human judgment, and warns that judgment is learned by doing the work AI would take over. Markus Brunnermeier has hosted several conversations about how AI can reshape economic research. And a recent dust-up over using AI to score points in an economic argument shows economists have a long road ahead in establishing norms on AI use.
Questions of mission, quality, pipelines, intellectual property, accountability, and tradeoffs are complex and will vary across institutions. Asking what AI can replace is an insufficient question. Asking what we accomplish with AI and people, now and in the future, is a better one. Don’t forget the juniors. They are not just affected by the answer; they are a big part of it.