For a decade, the standard worry about artificial intelligence was that it would hollow out white-collar work. In 2026, one of the most revealing trends in the economics job market points the other way: the frontier AI labs are competing, aggressively and expensively, for the best-trained economists of their generation.
This is not a vanity hire. The move toward increasingly general AI systems has stopped being purely an engineering problem and become a macroeconomic one. Labs that take their own forecasts seriously need to know what happens to labour markets, growth, inequality and public finances when the cost of cognitive work collapses. Engineers cannot answer that. Economists can.
The timing matters. While the academic market contracts, AI labs have quietly become prestige employers for senior economists, competing head-to-head with Ivy League departments and central banks for the same shortlist of names.
The profiles labs are buying
The clearest signal is the academic exodus. OpenAI appointed Duke professor Aaron "Ronnie" Chatterji as its first chief economist, and Anthropic advertises research economist positions inside a dedicated economic research team. These are not postdoc slots created to tick a box: labs are hiring tenured faculty, people who had already won the academic game and chose to leave anyway.
Three specialisations are in unusual demand:
1. Labour economics
Someone has to measure task-level substitution, retraining effects and whether the productivity gains firms report are real. Labs need credible numbers before regulators invent their own.
2. Game theory and mechanism design
As autonomous agents start transacting with each other, somebody has to design the incentives, pricing and matching rules they operate under. This is market design, applied to a market whose participants are software.
3. Macroeconomics and public economics
If capital substitutes for cognitive labour at scale, the questions become distributional and fiscal: who owns the surplus, what happens to the tax base, and what transfer systems survive contact with the result.
What the job actually looks like
Modelling scenarios almost nobody else models. What does growth theory say when labour supply becomes effectively unconstrained and nearly free? Most of the profession treats the question as speculative. Inside a lab, it is the planning assumption.
Access to data no university has. This is the honest reason many economists say yes. Labs hold telemetry and usage data on how millions of people actually delegate work to AI systems, at a granularity no survey will ever reach. That access is not available from the outside, at any price.
Policy and regulatory work. A large share of the role is translation: turning internal findings into evidence that governments, competition authorities and labour ministries can use, at a moment when both employment shocks and market concentration are live political issues.
Who is recruiting, and what each path costs
Most of these roles never reach a public listing, which is why the hidden job market matters more here than almost anywhere else.
Employer | What they hire for | The trade-off |
|---|---|---|
OpenAI, Anthropic | Named economic research teams attached to policy, governance and frontier-risk work | Publication is possible but the research agenda is set internally |
Google DeepMind, Meta | Behavioural economics, experimentation, monetisation of open agent ecosystems | Closer to product than to policy; less public-facing output |
Consultancies and policy shops | Interpreters between labs and public institutions | Client-driven work, shorter horizons, little proprietary data |
Central banks and IOs | Independent analysis of the same questions | Slower, structured hiring and far lower pay, with more independence |
The uncomfortable questions
Public research is being drained. Every senior labour economist who moves in-house is one fewer independent voice capable of evaluating these systems. If the people best equipped to audit AI's economic effects all work for the labs, independent evaluation gets harder.
The conflict of interest is structural. An economist publishing on AI and employment while employed by a company whose valuation depends on the answer is in an awkward position. Most labs permit publication; the question is which questions get asked in the first place.
The culture gap is real. Tech runs on weekly iteration, academic economics on three-year referee cycles. Researchers who move describe the speed as liberating at first and disorienting later, particularly when a finding ships as a blog post rather than a paper.
What this means for your career
Credible causal inference remains the core asset, but the differentiator is applying it to a domain with almost no historical data. Candidates who can read a model card, understand how an agent system is deployed and speak fluently to engineers are rare, and rarity is what the labs are paying for. If you are building that profile now, start with the skills employers are actually asking for in 2026.
Be clear-eyed about the bargain: pay competitive with finance rather than academia, a less predictable publication record, and a job that may not exist in its present form in five years.
The broader point is harder to miss. Labs hire economists when a technology stops being a research programme and starts being an economic institution. The next decade of the profession will not only be about studying human markets. It will be about designing and regulating economies in which a growing share of the participants are synthetic.
