When a company builds a bridge, it hires engineers. When a company starts worrying about where the bridge lands, who crosses it, and what happens to the towns on either side, it hires economists. Google just did the second thing.
In 2026, the company expanded its AI & Economy team, bringing in Nobel Laureate Philippe Aghion and Professor Ajay Agrawal, with new directors Anu Madgavkar and Daniel Rock leading the effort. The program’s stated focus is global AI adoption and labor market shifts. That’s a short announcement with a long shadow, and I think it’s worth unpacking for anyone who doesn’t follow economics papers for fun.
Why economists and not more AI researchers
Here’s what strikes me about this hire. Google already has an enormous number of people who understand how AI models work. What it apparently wanted more of was people who understand what happens after the models work.
Those are genuinely different questions. “Can an AI agent handle customer service tickets?” is an engineering question, and the answer is increasingly yes. “What happens to the 200 people who used to handle those tickets, the local economy where the call center sits, and the wage expectations of the next generation entering that field?” is an economics question. Nobody has a clean answer, and the answer will differ enormously depending on which country you’re standing in.
Aghion’s work sits in the tradition of studying how new technology drives growth over long stretches of time. Agrawal has spent years thinking about AI in business terms specifically. Madgavkar and Rock now direct the program. That’s a team assembled to measure something, not to build something.
What “labor market shifts” actually means for you
I want to translate that phrase, because it’s doing a lot of quiet work.
When researchers talk about labor market shifts from AI, they’re usually asking a handful of practical questions:
- Which specific tasks inside a job get automated, and which stay human
- Whether that leaves workers more productive and better paid, or simply replaceable
- How fast the change moves, because speed matters more than direction for real people
- Whether the gains concentrate in a few companies and countries or spread widely
Notice that none of these are answered by looking at a model’s benchmark scores. You answer them by looking at payroll data, job postings, productivity numbers, and survey responses across many economies. That’s slow, unglamorous work. It’s also the only way anyone will ever know whether the optimistic forecasts or the grim ones were closer to right.
The global adoption angle is the underrated part
The program’s focus on global AI adoption deserves more attention than it will probably get. Most conversation about AI and work assumes a wealthy-country context: knowledge workers, laptops, software subscriptions. But adoption patterns are going to look wildly different in economies where the dominant employers are agriculture, manufacturing, or informal services.
An AI agent that drafts legal memos changes one kind of economy. An AI tool that improves crop yield predictions or translates technical documentation into a dozen languages changes a different kind entirely. Studying adoption globally means accepting that there isn’t one story here. There are dozens, and some of them are happening in places that rarely appear in tech coverage.
Reading the move with clear eyes
I’ll be straightforward about the obvious tension. Google builds and sells AI. A research program it funds, studying the economic effects of the thing it sells, is not the same as independent academic work. The credibility of the people involved helps, and hiring a Nobel Laureate signals you’re not looking for a rubber stamp. But the incentive structure is what it is, and readers should hold the findings to the same standard they’d apply to any industry-funded research: check the methods, check whether the data is available to others, check whether inconvenient results ever see daylight.
That said, I’d rather this research exist than not. The alternative to well-resourced study of AI’s economic effects isn’t neutral silence. It’s everyone arguing from anecdote and vibes, which is roughly where we’ve been for a while.
What I’d watch next
The useful signal won’t be the hiring announcement. It’ll be what the program publishes, how specific the findings get, and whether the work engages with uncomfortable results as readily as encouraging ones.
For those of us trying to explain AI agents to people whose jobs might change because of them, better data would be a real gift. Right now the honest answer to “will this affect my work?” is usually “probably, in ways nobody has measured yet.” Teams like this one exist to shrink that second half of the sentence. That’s worth paying some attention to.
🕒 Published: