Remember when every AI company’s answer to “will this take my job” was a blog post with a stock photo of a smiling worker and a chatbot? That was the standard playbook for a good while. Vague reassurance, no numbers, no methodology, and a link to a free course.
Google is trying something different. In 2026 the company expanded its AI & Economy team, bringing in Nobel Laureate Philippe Aghion and Professor Ajay Agrawal. New directors Anu Madgavkar and Daniel Rock are running the show. The AI & Economy Research Program is aimed at global AI adoption and labor market shifts. That is the whole confirmed picture, and I want to be straight about that before I offer an opinion on it.
Why a toolkit reviewer cares about economists
I spend my days testing agent frameworks, coding assistants, and workflow automation tools. I write up what works and what breaks. The thing I can tell you whether a tool saves you twenty minutes on a pull request.
That gap bothers me. Every review I publish implicitly assumes the tool is a productivity multiplier for the person using it. That assumption is doing a lot of work, and it has never been properly tested at scale. Aghion has spent his career on growth and creative destruction. Agrawal has written extensively on the economics of prediction. Rock’s work sits at the intersection of AI and labor. Madgavkar comes from the research side of understanding how work actually changes. These are people who study exactly the question I keep dodging.
The obvious caveat
Google is paying for this. Any research program funded by a company with a direct commercial interest in AI adoption carries that weight. I am not going to pretend otherwise, and neither should you when the first papers land.
That said, the alternative is not neutral research. The alternative is no research, or research funded by whoever else has money. Academic economists with established reputations tend to guard those reputations carefully, because the reputation is the whole asset. A Nobel Laureate does not need a Google affiliation badly enough to produce findings that get torn apart at the next conference. The incentive structure is imperfect but not empty.
What I want out of this
Here is my honest wish list, as someone who reviews tools and has to keep a straight face about the word “productivity”:
- Adoption data that separates trial from actual sustained use. The gap between “we deployed Copilot” and “our engineers use Copilot daily six months later” is enormous and almost nobody measures it.
- Task-level breakdowns rather than job-level ones. Jobs do not disappear, tasks get reassigned. I need to know which tasks.
- Country and sector variation. Global adoption is in the program’s stated scope, and I hope that means more than a US-plus-Europe sample.
- Negative results. Tools that got rolled out and did nothing. Those are the reviews nobody writes and everybody needs.
The last one matters most to me. My review inbox is full of pitches promising transformation. My testing notes are full of tools that were fine, did roughly what the docs said, and changed nothing about how a team operated. Somebody with actual data and actual methodology should be writing that finding down.
What this signals about the broader space
Hiring economists is not a marketing move in the usual sense. It does not produce a demo, a launch video, or a benchmark score. It produces papers that take a year to write and get argued about for three more. Companies do that when they expect regulatory and public scrutiny to get serious, and they want a seat at the table with something other than vibes.
For those of us evaluating tools, that shift is useful even if we never read a single paper. It means the conversation is moving from “look what this can do” to “what does this actually change.” Those are different questions, and the second one is harder and more useful.
My take
I am cautiously positive and holding judgment on the output. The names are strong, the stated scope is the right scope, and the funding source deserves scrutiny rather than dismissal.
What I will do is read the first outputs and hold them to the same standard I hold a tool review. Does the method match the claim? Are the limitations stated plainly? Would a skeptical reader come away better informed or just better marketed to? If the answer to those is yes, I will link to the work and recalibrate how I write about productivity claims. If it is no, I will say that too.
🕒 Published: