OpenAI's own economists cannot scope their jobs a year out

Roughly two months of shelf life ^[1] is the number that governs everything else about how this team is built.

An economics function whose evidence ages that fast cannot run on annual work plans, and cannot be recruited against a fixed brief. So the brief becomes temperament. Chatterji says he needs rigorous economic thinking applied to what is happening now, together with willingness to take on questions that have no settled answers, or in his phrasing, people who are "comfortable with being uncomfortable" ^[7]. He also says he is hiring more people and declined to say how many ^[5], which leaves the interesting quantity unpublished: the distance between what the team has been asked to cover and what it can actually staff.

The original architecture was three questions, according to Chatterji: how AI is changing work today, how businesses are adopting and reorganising around it, and what more capable AI could mean for the economy later ^[6]. Recursive self-improvement is a topic that arrived after those three were set ^[2]. That is the shape of the problem for anyone else hiring in this area. The founding questions were not wrong. They were simply not exhaustive, and they stopped being exhaustive inside a single planning year.

Where the role sits is worth more attention than org charts usually deserve. Chatterji reports to OpenAI's finance chief, Sarah Friar ^[10], not into a research or policy line. The job itself was reverse-engineered from a conversation. He joined in 2024 after coordinating the $52 billion CHIPS program and serving as acting deputy director of the National Economic Council ^[11], and the discussions that brought him in began on supply chains and semiconductors before widening until the chief economist role was created around them ^[12]. A function invented that way has no institutional precedent to fall back on when the questions move, which is one reason the rescoping shows up in the job description rather than in a quiet internal reshuffle.

Chatterji also describes the pace of innovation at the company as "a little insane" ^[9], and puts the demand for self-direction in concrete terms: team members are expected to work out which problems are worth solving and who the output is for ^[14]. Those two statements sit together. Speed at the model layer is what pushes the scoping decision down to the individual researcher, because nobody above them can hold a stable list.

For anyone writing an AI-impact requisition, the useful read is not OpenAI's topic list. It is that a team with a direct line to the model roadmap is still rescoping itself mid-year, while the people importing its published findings into their own hiring plans are working from a fixed brief and an ageing dataset.