A few years ago, the title "chief AI economist" did not exist. Today it sits inside one of the largest technology companies in the world, and that alone says something about where the analytical center of gravity has moved.

A chief economist reads the macro conditions outside a firm against the micro conditions inside it and turns that reading into strategy. Adding "AI" to the title changes the scope. AI is no longer one input among many in a cost structure. It is becoming the medium through which labor, capital, productivity, and demand are being re-sorted, which is a large enough shift to justify a dedicated seat.

The work done from that seat is also increasingly outward facing. At Microsoft, the AI Economy Institute sits alongside the AI for Good Lab, which has spent seven or eight years applying data and models to earthquakes, wildfires, deforestation, and agriculture in developing contexts. The Institute extends that mandate into economics, funding academics as fellows and asking them to write for a general audience as well as for their peers. That structure answers a real problem: most of what gets said about AI and the economy is said quickly, from intuition about what the technology appears able to do.

Why the researchers are moving

The story of academics joining technology companies is usually told as a loss. It is better understood as a story about proximity.

Academia offers something industry structurally cannot, which is the time and space to go deep on a question until you actually understand it. What has changed is that the object of study now moves faster than the outside view can track. If you want to know how a technology is being built, what is coming next, and how people are actually using it, a growing share of the signal sits inside firms. Partnership from the outside helps. It does not close the gap.

At micro1, we saw this while recruiting for the Economics Research Unit. Candidates from the top doctoral programs, with real academic options, chose a technology company. I crossed that bridge later in my own career, once it was clear how little of the work I cared about could reach a wide audience from inside an institution. Their calculation starts earlier, and it is not naive. A university postdoc gives you new colleagues and new ideas. A lab gives you those plus data and a working view of the technology.

What makes the arrangement healthy is independence. Funding a researcher is not commissioning one. Help with writing and framing is legitimate; editorial control over findings is not.

Collaborate on the core, compete on the edges

That is the title of Frank Nagle's article in the September print edition of Harvard Business Review, and it comes out of a long research program on open source. The original framing tied it to AI. HBR broadened it, correctly, because the argument travels much further.

The puzzle is genuine. By conventional strategy logic, paying your own engineers to write code your competitors can use for free is indefensible. Firms do it anyway, and more every year, alongside research partnerships, standards work, and informal arrangements that competitive analysis struggles to explain. The article offers five factors for deciding when to collaborate and when to compete, based on a firm's position, its industry, the technology, and the maturity of the ecosystem around it.

One factor is where you sit in the technology's life cycle, and it maps onto AI almost too neatly. The idea has existed since the 1960s, but broad public use arrived only with generative models. We are early, and early is when collaboration dominates. Hence interoperability standards such as MCP and A2A developed in the open, and hence traditional competitors working together to help regulators understand where the technology is heading.

The same logic applies to the open versus closed debate, which is usually conducted as a moral argument and is better read as an economic one. Expect specialization rather than victory. Operating systems are the precedent: closed came to dominate the consumer desktop, open came to dominate the server. Token maxing, the reflex of spending as much as possible on the most capable model for every task, is already giving way to routing work by what the task actually needs.

The labor question is a transition question

The predictions of mass AI-driven unemployment have not materialized in the data. That is worth saying plainly, because it is good news and because it is usually skipped.

What the data does show is turbulence in specific places: entry-level roles, in particular sectors, doing the kinds of tasks AI performs efficiently. Whether that stays localized or is the leading edge of something broader is an open question, and anyone who claims to know does not.

The right historical reference is not the Luddites but the China shock. When China joined the WTO, developed economies understood that manufacturing was going to move. The shift was foreseen. The transition was still managed badly, with too little retraining and too little support, and we are still living with the consequences. Foresight without preparation is worth very little.

So the labor problem is really an identification problem. Where is automation landing, who is exposed, and what is the smallest intervention that moves someone into work less likely to be disrupted? Not five years of retraining, but short and targeted programs. There is a related trap on the cost side: substituting cheap models for entry-level work looks obvious in the short run, until a firm discovers it has delegated judgment it no longer has anyone in-house to reconstitute.

What we cannot measure, we will not manage

If we cannot measure something, we cannot improve it, and our headline measures were not built for this. Simon Kuznets, one of the architects of national accounting, warned against using GDP the way we now use it. Compressing an economy into one number was always a compromise, and it strains under technologies whose price is zero. Open source is the clean case: enormous value, almost no direct footprint in the statistics.

This is where I am more optimistic than most people I talk to. Weak measured productivity against heavy AI investment is not evidence of failure, since general purpose technologies have never paid off on quarterly cycles. It is evidence that our instruments are too coarse. The pressure AI puts on short-term measurement may finally push us toward a broader taxonomy of metrics.

Diffusion needs the same scrutiny. Electricity took decades to reach most of the world. The adoption statistics for AI are remarkable by comparison and also partial, describing the United States, Western Europe, and a handful of comparable economies. The encouraging counterexamples are unglamorous. Teams working in Kenya describe people reaching AI over text message, with a nonprofit using that channel to put basic medical knowledge within reach of people far from any clinic. The question a mother needs answered is often just whether what she is experiencing is normal or warrants seeing a professional. Hallucination and misinformation are serious risks there and have to be engineered against. But an interface that works on a feature phone is a real answer to a real access problem.

The question worth working on

Asked what he most wants to investigate from his new seat, Frank Nagle gave two answers that turn out to be one: how to keep the technology more augmentative than automating, and how to keep its benefits diffuse rather than concentrated in a few firms, countries, and individuals.

Both are questions about distribution rather than capability. Capability is being solved, expensively and fast, by a large number of very good people. Distribution is not on the same trajectory and will not fix itself. That is why the short term deserves vigilance. Optimism about the long-run equilibrium is defensible; using it as a reason to neglect the present undermines the foundation the long run has to be built on.

This piece draws on a conversation hosted by the micro1 Forum, featuring Frank Nagle, Chief AI Economist at Microsoft and Research Scientist at MIT, in conversation with Mark Esposito, Chief Economist at micro1.