Strip away the announcement gloss and the job handed to Anu Madgavkar and Daniel Rock sounds almost modest. As the new directors of Google’s AI & Economy team, their brief is to track AI’s influence on jobs, productivity, and global markets. That’s it. No model to ship, no benchmark to beat. Just: figure out what this technology is actually doing to the economy people live in.
My reaction, as someone who spends her days explaining AI agents to people who never asked for them: finally, somebody is being paid to measure the part that matters to the rest of us.
Who Google brought in
The 2026 expansion added Nobel Laureate Philippe Aghion, Professor Ajay Agrawal, and other leading researchers to the team. That’s an economist roster, not an engineering one. Nobody on that list is optimizing inference costs.
If you’ve spent the last few years watching AI news, you’ll notice how unusual that is. The headlines have mostly been about capability: bigger models, faster chips, agents that can book your travel or refactor your codebase. Google’s own hardware news this year followed exactly that pattern. Capability is easy to demo and easy to write about.
Economic impact is neither. It shows up slowly, unevenly, and in places nobody was watching. Which is presumably why you need Nobel Laureates to find it.
The three things they’re actually measuring
The team’s stated focus breaks into three pieces, and each one means something specific if you translate it out of research-speak.
Jobs
Not “will AI take my job,” which is the question everyone asks and nobody can answer cleanly. The researchable version is narrower: which tasks inside which jobs are shifting, who’s doing more of them, who’s doing fewer, and what happens to pay. An AI agent that drafts your emails doesn’t eliminate your role. It changes what your day is made of. Those changes accumulate into something measurable, but only if somebody is keeping count.
Productivity
This is the quiet one, and the one economists care about most. Productivity means output per hour worked. Every major technology of the last century has been sold on the promise of raising it, and the historical record is messier than the marketing. Electricity took decades to show up in factory numbers. Computers had their own long lag. If AI agents are genuinely making work faster, that should eventually appear in national statistics, and if it doesn’t, that’s a finding too.
Global markets
Impact won’t distribute evenly across countries. Economies built on the kind of work AI handles well will feel this differently than economies built on the kind it can’t touch. Tracking that is less about technology and more about trade, wages, and who owns the infrastructure.
The obvious tension
Let’s name it: Google is a company with enormous financial stakes in AI adoption, and it is now funding research into whether AI adoption is good for the economy. That’s a structural conflict, and pretending otherwise would be silly.
Two things soften it. Academic researchers with established reputations have their own incentives, and those incentives point toward publishing what they find rather than what their host prefers. And the alternative isn’t neutral either. Without corporate-funded research, the measurement work mostly doesn’t happen, or happens years late with worse data access.
The useful posture is neither dismissal nor trust. Read what this team publishes, check who peer-reviewed it, and compare it against independent work. That’s how you read any industry-adjacent research.
Why this matters if you’re not an economist
Here’s what I think non-technical readers should take from this. Right now, conversations about AI agents at work run on vibes. Someone in your office says agents will handle half the support queue by next year. Someone else says it’s all hype. Neither has numbers, so the loudest person wins.
Research like this eventually gives you something better than vibes:
- Actual figures on which tasks agents are absorbing, so you can judge whether your role is in that set
- Evidence on whether productivity gains are real or just shifted work, which matters when your employer justifies a change
- Data that policymakers can use, instead of anecdotes from whoever briefed them last
None of that arrives next month. Economic research is slow by nature, and the honest version of this story is that the first solid findings are probably years out.
Still, the signal is worth reading. When one of the largest AI companies decides it needs Nobel-level economists on staff, it’s an admission that nobody currently knows the answer to the question everybody’s asking. That’s a more honest starting point than most AI announcements give us.
đź•’ Published: