If you have ever tried to justify an AI feature with a number, you know the problem: usage data tells you what people clicked, not what changed for them. Google’s AI & Economy program is a bet that the second question needs economists, not just dashboards.
What Google announced on September 18, 2026
In a post on the Google blog, Scott Strand and Zanna Iscenko described an expansion of the company’s AI & Economy Research Program, adding external advisors, visiting scholars, and two research directors. The program follows the earlier launch of AI & Economy ATLAS v1.0 and its open-access site, which track how people use Google’s AI tools at work and in daily life. Google frames adoption tracking as a starting point, not the goal.
The named additions:
- Philippe Aghion joins as an Academic Advisor, alongside Michael Spence and Dame Diane Coyle, bringing work on innovation-led growth to questions about AI’s long-run macroeconomic path.
- Ajay Agrawal joins as a Visiting Fellow, working with David Autor on the economics of AI and scientific discovery, AI and robotics, and human welfare.
- Anu Madgavkar, formerly of the McKinsey Global Institute, and Daniel Rock of the Wharton School become Directors of the research program. Madgavkar will lead empirical work on global AI diffusion, small business ecosystems, and workforce impacts; Rock will connect frontier model telemetry with econometric analysis of enterprise productivity and labor restructuring.
Madgavkar and Rock will lead the program alongside Alex Imas, Director of AGI Economics at Google DeepMind, and Iscenko in Google’s Chief Economist’s Office.
Why the telemetry-plus-econometrics pairing matters
The detail worth pausing on is Rock’s brief: bridging frontier model telemetry with rigorous econometrics. Most product teams already sit on a version of that telemetry — token counts, session depth, task completion, retention by cohort. What they usually lack is a credible way to convert it into claims about productivity or labor impact that a CFO, a regulator, or a customer’s procurement team will accept.
That gap is not unique to Google. It shows up whenever a builder has to defend an AI feature on outcomes rather than novelty. The same question runs through what dual jurisdiction AI procurement means for your build: buyers increasingly want evidence about where the system runs and what it changes, not just a capability list.
Google says the expanded team will directly inform future ATLAS updates and empirical research, with focus areas including the future of work, productivity and growth, global technology diffusion, and AI’s impact on scientific discovery. The stated aim is to identify organizational practices, policy frameworks, and training programs that help AI upskill workers and spread expertise.
What is confirmed, and what is not
Confirmed: the appointments, the program’s research areas, and the link between the team and ATLAS updates, all as described in the Google post dated September 18, 2026.
Not confirmed: any specific methodology, dataset, publication schedule, or metric definitions. The post does not say how telemetry will be joined to economic data, what access outside researchers will get, or when findings will land. If you are waiting on a citable productivity benchmark to put in a business case, this announcement is a signal about direction, not a deliverable.
A practical read for builders
Treat this as a slow-moving input, not a launch. Two things are reasonable to do now. First, keep your own instrumentation honest about outcomes, not just engagement, so you are ready if buyers start asking for evidence in Google’s framing. Second, watch the ATLAS updates as they arrive — they are one of the few public windows into how a major platform measures real usage of AI at work.
The limitation is timing. Research programs of this shape tend to publish on academic cycles, and the post gives no dates. Plan around your own measurements in the meantime.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.