Mike Hunstad runs Northern Trust's $1.4 trillion asset management division. In a recent interview with the Financial Times, he made a claim that has been picked up approvingly across the financial press: AI is going to be "massively disinflationary," and the Federal Reserve should hold rates steady while it waits for the productivity gains to show up.

The first part of that argument is the kind of thing reasonable people can disagree about. The second part is a category error.

Hunstad's framing — "it's almost like AI is your monetary policy, and it's going to be more effective than anything the Fed or really any central bank around the world can do" — treats AI as a policy lever. It is not a lever. It is private investment that may or may not, eventually, become productivity. Conflating the two is how you get the most expensive wait-and-see posture available to a corporate decision-maker, and the most consequential one.

I have no interest in disputing the underlying productivity case. The microeconomic evidence is real. Coders complete certain tasks meaningfully faster with AI assistance. Customer support agents close more tickets and escalate fewer of them. Knowledge workers can compress an hour of writing into a fraction of that. Customer-facing teams given predictive customer intelligence intervene earlier and on the accounts that actually matter. None of this is in serious dispute. The question is what happens in the long stretch between those task-level gains and an economy-wide productivity number — and how confident anyone should be about the timing.

The honest macro answer is closer to the one Daron Acemoglu has been giving. As a recovering economist, I will admit that the dismal science earned its nickname — and that the dismal version of the answer is usually the more reliable one. Total factor productivity gains over a decade in the range of half a percent to two thirds of a percent. Goldman Sachs' optimistic case, around eight percent over the same horizon. These are useful numbers. They are also a long way from "more effective than anything the Fed or really any central bank around the world can do."

Investment is not productivity

The deeper problem with Hunstad's argument is one his Fed counterpart already named, and which the same article quietly carried in a single paragraph.

Philip Jefferson, the Fed's vice chair, made the point that "soaring investment in AI infrastructure, such as data centers, boosts demand immediately — and risks raising prices, while the effects on productivity will take longer to play out." That is the correct reading. When investment runs visibly ahead of productivity, economic historians have a few words for what comes next. Disinflation is not on the list.

The economy is currently in the spending phase of the cycle. The trillion-dollar capital expenditure is real and concentrated and now. The productivity, if it arrives at scale, arrives after a re-composition of how work is organized — and re-composition, every time we have studied it, has taken a decade or longer.

The PC was visible everywhere except in the productivity statistics for fifteen years. Electrification took thirty. The first commercial railway opened in 1830, and people were still arguing in 1850 that the whole thing had been overhyped. Each of those technologies was eventually more transformative than its early advocates predicted. None of them was disinflationary on the timeline anyone expected at the moment of investment, because the bill arrived first and the productivity arrived second. Productivity arrives in a pattern that economic historians find consistent and that quarterly earnings calls find inconvenient.

Warsh's favourite comparison is the 1990s productivity boom — which took fourteen years to show up in the data after the PC arrived. He means it as encouragement. The reason that boom took fourteen years is the reason it always takes that long: technology arrives before organizations are reorganized to use it. Warsh, without intending to, is conceding exactly the gap that Jefferson is describing.

The asset manager and the operator are not playing the same game

It is a curious feature of capital allocation that the people most confident about productivity arriving on schedule are also the people who do not have to deliver any of it themselves.

Asset managers can wait. Their portfolio absorbs the time lag. The cost of being early is low. The cost of being right late is approximately zero. If the productivity arrives in 2032 instead of 2026, the asset manager writes a slightly different annual letter.

The operator writes a press release explaining the layoffs. The cost of waiting, for a particular company, is the difference between being one of the firms that has rebuilt its work around AI and one of the firms that has not — and that difference compounds. If your competitor spent 2026 quietly redesigning how customer intelligence flows through their organization, and you spent 2026 holding steady to see what the true AI productivity benefits entail, the gap between the two of you in 2030 will not be small. It will be structural, and it will be very difficult to close from behind.

Asset managers are paid in basis points to be patient. Operators are paid in earnings to be on time. These two compensation structures rarely produce the same advice.

The wait-and-see posture is the most expensive one available

Hunstad's prescription to the Fed is that it should "hold steady and put a communication out to the market that, 'hey, we're just going to hold steady and see what the true AI productivity benefits entail.'" The phrase "true AI productivity benefits" is doing a remarkable amount of work in that sentence.

What he is describing is patience as a strategy — Patience-as-a-Service, presumably next in line after the rest. Wait for the data to come in. Resist the urge to act prematurely. Let the productivity reveal itself. For a central banker, this is a defensible posture; central banks are designed to lag, and the cost of premature easing is real. For an operator running a customer-facing business, the same posture is a slow-motion mistake.

It is a feature of large pools of capital that they can outwait almost anything. It is a feature of operating businesses that they cannot.

The companies that will look prescient in 2030 are not the ones that waited for the productivity benefits to be measurable in the aggregate before deciding to act. They are the ones that decided, before the picture was clear, which parts of how they ran their customer operation were tacit pattern recognition that an AI system could now do better than the people doing it today — and quietly began rebuilding the work around that capability while their competitors waited for permission. The data telling them they were right will arrive after the advantage is locked in. By the time the productivity benefits are visible to a central banker, they will have already been distributed unevenly, and the distribution will look like which firms acted and which firms waited.

What this means for the operator

The AI productivity gains will arrive. Whether they are massively disinflationary or modestly disinflationary or simply distributed unevenly across industries is a question for economists and central bankers, and they can argue about it with my blessing. None of it changes what an operator should be doing now.

The operating question is not "when will AI lower my cost base." It is "which parts of how I currently run my customer operation are tacit pattern recognition that an AI system can now do better than the people who are doing it today, and what does the organization look like when those parts are reorganized around the new capability." That is not a wait-and-see question. It is a redesign question, and redesign happens on a calendar that has nothing to do with the Federal Reserve's.

The temptation of Hunstad's framing is that it lets an operator outsource a decision they should be making themselves. If AI is monetary policy, then someone in Washington is in charge of getting the productivity to your P&L. You can stand down. The reality is the opposite. The productivity, when it arrives in your particular company, will arrive because someone in your company decided to reorganize the work — or it will not arrive, because nobody did. There is no central bank for that decision. There is only you.

Sources: Harriet Clarfelt, Kate Duguid and Claire Jones, "AI boom poised to be 'massively disinflationary', Northern Trust says," Financial Times, 19 April 2026; Daron Acemoglu, "The Simple Macroeconomics of AI" (NBER, 2024); Goldman Sachs Global Economics Research; Bureau of Labor Statistics productivity data, 1981–1995.