Hiring economists is not a sign that a company is being careful. It is often a sign that a company knows a fight is coming and wants the referees on payroll before the bell rings.
That is my read on Google’s 2026 expansion of its AI & Economy team, which now includes Nobel Laureate Philippe Aghion, Professor Ajay Agrawal, and other researchers brought in to study AI’s global economic impact. New directors Anu Madgavkar and Daniel Rock are running the show, with a stated focus on AI adoption and labor market shifts. The broader AI & Economy Research Program exists, per Google, to advance understanding of how AI is reshaping economies.
The mainstream take is that this is good news. Serious academics studying serious questions, funded by a company with the data to answer them. And honestly? Parts of that are true. But I review AI tools for a living, and the first question I ask about any product is who benefits when the results come back favorable. That question does not stop applying just because the output is a working paper instead of an API.
Why this roster actually matters
Let me give credit where it is due, because this is not a vanity hire. Aghion’s work on growth and creative destruction is exactly the lens you want when the question is whether AI expands output or just reshuffles who captures it. Agrawal has spent years on the economics of prediction machines, which is a less glamorous but more useful framing than most of what gets posted about AI on a given Tuesday. Rock has done some of the more grounded empirical work on which occupations are actually exposed to language models rather than which ones sound exposed. Madgavkar brings the labor market side.
These are not people who need Google to be relevant. That is the part worth paying attention to. If you wanted a team to produce flattering slides, you would not pick researchers whose reputations depend on being wrong in public as rarely as possible.
The structural problem nobody wants to name
And yet. The most important research questions about AI and the economy are questions where Google is a defendant, not a neutral party.
- Does AI adoption raise wages or compress them? Google sells the adoption.
- Which jobs get automated first, and who eats the cost? Google builds the automation.
- How much of the productivity gain flows to workers versus to the companies that own the models? Google owns models.
- What should regulators do about labor displacement? Google has a very specific preference about the answer.
None of that means the research will be dishonest. Good economists working in industry produce good work all the time. But funding shapes which questions get asked, which datasets get shared, which findings get a press push, and which get released quietly on a Friday. That is not a conspiracy. It is how institutions behave, and pretending otherwise is naive in a way that helps nobody.
The tell will be in the uncomfortable findings. If this program publishes work showing that AI adoption in a specific sector reduced headcount without raising output, and Google amplifies it as loudly as it amplifies the optimistic stuff, I will update my view fast. If the only results that get a blog post are the ones about augmentation and new job categories, we will know what this is.
What I want to see
For a program like this to be more than a reputational asset, a few things need to be true, and they are all checkable over time.
- Data access that outsiders can use. Google sits on adoption telemetry no university can replicate. Sharing conclusions is nice. Sharing data that lets rivals test those conclusions is the actual commitment.
- Preregistration. If you say what you are measuring before you measure it, you cannot quietly drop the analyses that went badly.
- Publication independence in writing. Not vibes. Not assurances. Terms that let a researcher publish a finding the sponsor hates.
- Negative results on the record. The AI economics literature is already thick with optimistic projections. What is thin is careful documentation of where AI did not deliver.
My actual position
I am cautiously positive, which for me is a strong endorsement. The AI economics conversation right now is dominated by vendor forecasts and consultancy decks with numbers that fall apart the moment you ask for methodology. Putting Aghion, Agrawal, Rock, and Madgavkar on the problem raises the floor on quality, even accounting for the funding source. Better to have sharp people asking the right questions under a conflict of interest than mediocre people asking the wrong ones under none.
Just do not confuse a research program with a commitment. Google has funded plenty of work on AI’s economic effects while shipping products that create those effects at speed. Both things can be sincere. Only one of them is measurable by revenue, and that asymmetry does not vanish because the letterhead now carries a Nobel.
Read what this team publishes. Read it carefully. And notice what they choose not to study.
🕒 Published: