“Sometimes unlearning is harder than learning.”
Last year on August 18th, with AI bubble fears still high, before software had collapsed, before DRAM prices were rising, and with PMIs still below 50, I published a paper titled “The Academic Fed vs. the Inflation Target of the Future.” The last paragraph read:
“The era of the ‘academic Fed,’ with its rigid devotion to outdated models and a 2% inflation target born in the 1990s, is giving way to an ‘inflation target of the future,’ one shaped by fiscal expansion, technological acceleration, and the realities of exponential innovation. Just as railroads, industrialization, and the IT revolution redefined markets in past centuries, today’s alignment of fiscal, monetary, and technological forces marks the start of a secular regime shift unlike any before it. AI is dismantling the moats of software incumbents while reviving demand for physical infrastructure, energy, and hard assets that have been underinvested in for nearly two decades. For investors, the implication is clear: the winners of the next era will not be those who cling to yesterday’s orthodoxy, but those who recognize that higher inflation tolerance, productivity-driven growth, and a capital rotation toward the physical world are reshaping the opportunity set. Those who adapt early will capture the upside of this transformation, while those bound to the past risk being left behind.”
The inspiration for that paper was an interview with Scott Bessent on the All-In Podcast. This was one of the key paragraphs referencing that interview:
“Bessent’s belief that AI-driven productivity shifts demand a forward-looking Fed has real merit. AI is progressing at an unprecedented pace, and in a world defined by exponential innovation, relying on backward-looking models is too slow. As digital employees begin to replace human workers across industries, deflationary pressures from automation will be powerful and persistent. The labor vs capital imbalance will likely worsen. This creates a paradox: the labor market may weaken even as overall economic output and profitability remain strong, fueled by productivity gains. In such a regime, clinging rigidly to a 2% inflation target risks policy mistakes, whereas embracing exponential innovation allows the Fed to harness growth without unnecessarily stifling it.”
That paper was written before the market fully accepted what AI was doing to software, before the physical infrastructure boom was obvious to everyone, and before a new Fed chair was openly talking about AI as a central banking paradigm shift. Over the last year, investors have been forced to learn AI the hard way. After two years of calling it a bubble, dismissing it because of hallucinations, and waiting for the hype to fade, AI instead consumed global equity markets and began reshaping economic reality. The software sector has already felt this repricing. Investors trying to pick the bottom in disrupted software names learned that the market discounts structural change before it shows up cleanly in reported numbers.
Now the same lesson is coming for economists and Fed watchers. Many of them have spent the past year minimizing AI’s macro importance, treating it as too speculative, too early, or too hard to measure. But that is exactly the point. The economy is changing before the old data architecture can fully capture it. The first Fed chair of the AI age is taking over just as the global economy is beginning to absorb a new class of labor: digital employees. AI agents, copilots, and autonomous workflows are moving from novelty to production, with powerful new models and capabilities arriving weekly.
That means the next phase of Fed watching will require something harder than learning a new framework. It will require unlearning an inherited one. The critique I made last year about the “academic Fed” now applies just as much to the people analyzing the Fed. For decades, investors, economists, and market commentators have listened to the Fed through a familiar policy lens: hawkish or dovish, restrictive or accommodative, higher for longer or pivot, dots or no dots. That language worked in a more linear economy where policy, inflation, labor markets, and productivity moved with long and variable lags. AI is making that world less stable, less measurable, and more reflexive.
Just as software investors had to stop looking backward and start discounting disruption ahead of the reported data, Fed watchers will need to do the same with central banking. The old playbook of parsing every Fed phrase for a rate-hike or rate-cut signal is no longer enough. In a world where billions of digital workers may be entering the production function, the more important question is whether the Fed, and those interpreting it, can adapt fast enough to an economy being rewritten in real time.
“This is the most disruptive moment in modern economic history in the U.S. and the world.”
Kevin Warsh’s confirmation hearing and last week’s ECB Forum interview should be read together because they tell one consistent story. Warsh believes AI is more than another macro variable. He sees it as a force powerful enough to require a new central banking paradigm. At his confirmation hearing, he described the current moment as “the most disruptive moment in modern economic history in the U.S. and the world.” At the ECB Forum, he sharpened the same idea: “This is a big paradigm shift both for the conduct of our policy and for our economies.”
Those are not casual comments. They are the foundation of a new monetary-policy worldview. The academic history of monetary policy does not include AI agents, billions of software workers entering the global production function, or general-purpose intelligence being deployed across coding, customer service, research, finance, compliance, sales, logistics, and operations at near-zero marginal cost. If the economy itself is changing, then the way we listen to the Fed must change too.
“Status quo practices and policies are especially harmful when the world is changing this fast.”
The key bridge between the hearing and the ECB Forum is Warsh’s attack on institutional inertia. At the hearing, he warned against Milton Friedman’s “tyranny of the status quo,” saying old practices become especially harmful when the world is changing quickly. At the ECB Forum, he translated that into central banking language: the Fed needs to go back to first principles, rethink models, rethink communications, rethink data, and rethink the balance sheet.
The same warning applies to investors. The old Fed-watching playbook is also a status quo practice. The obsession with whether one press conference sounded restrictive or supportive assumes the economic structure is stable enough that words can be mapped directly into a rate path. In an AI economy, that assumption becomes dangerous. A Fed chair can defend inflation credibility while also recognizing that productivity is improving. He can be cautious on rates while believing the supply side of the economy is undergoing a profound transformation. The binary labels are too small for the moment.
“One is the increase in capital expenditures to build data centers and the rest that will have an effect on demand.”
Warsh’s cleanest AI framework is the demand-versus-supply puzzle. AI infrastructure spending is clearly a demand impulse. Data centers, chips, power, cooling, construction, electrical equipment, transmission, and labor all represent real capital spending. At the hearing, Warsh said the increase in capital expenditures “to build data centers and the rest” would affect demand and could increase demand by “a few tenths of 1%.”
He immediately separated that from the much larger supply-side possibility: “But on the supply side of the economy, to increase the potential output of the economy, that could be considerably bigger.” This is the nuance that most Fed commentary misses. AI capex may be modestly inflationary through demand today, while AI diffusion may be meaningfully disinflationary through supply tomorrow. Data centers are not merely buildings filled with chips. They are the physical infrastructure for a new productive layer of the economy built on AI factories and tokens.
“But on the supply side of the economy, to increase the potential output of the economy, that could be considerably bigger.”
This is where Warsh begins to sound like a modern Alan Greenspan. Greenspan’s great insight in the late 1990s was not that every internet stock was justified. It was that information technology was improving productivity, reducing uncertainty, and suppressing unit labor costs before conventional models fully captured the shift. Warsh is making a similar argument, but with a more powerful and faster-moving technology.
AI starts as a technology story, then becomes a capital spending story, then becomes a productivity story, and eventually disappears into ordinary business. At the hearing, Warsh put it this way: “Today we call it artificial intelligence. Two years from now we’re going to call it business capex and three years from now we’re going to call it just ordinary business.” That is exactly how general-purpose technologies diffuse. First the market debates the hype. Then companies reorganize around the tool. Then the tool becomes the production function. For the Fed, the question is whether potential output is rising faster than the official data can see.
“I am more confident that there will be improved output than I am certain about when the effects of that would be on the labor market.”
The labor-market question is even harder. Warsh is not arguing that AI’s employment effects are simple or painless. He is arguing that output gains may arrive before the labor-market effects are clear. At the ECB Forum, he rejected the doomer version of technological change and invoked the lump-of-labor fallacy, arguing that new technologies historically create jobs that were unimaginable at the start of the transition.
At the hearing, he added the central-banking complication: “I am more confident that there will be improved output than I am certain about when the effects of that would be on the labor market.” He also said, “The lag between the improvement in output and the effect on the labor markets, that’s got to be central to the Fed’s thinking given the pace of innovation in this cycle.” That is the Fed’s employment mandate in the AI age. Productivity can rise first. Hiring patterns can change second. Labor stress can appear unevenly across age cohorts, industries, and skill levels. The official unemployment rate may look stable while entry-level white-collar work is quietly being rewritten.
“The lag between the improvement in output and the effect on the labor markets, that’s got to be central to the Fed’s thinking.”
This is why the traditional rate-cycle shorthand becomes increasingly misleading. If AI raises output before it shows up in employment, the Fed may see stronger growth without the same wage pressure that would have appeared in the old economy. If AI substitutes for some tasks before it creates new job categories, full-employment statistics may hide significant internal displacement. If data centers raise near-term demand while agents raise long-term supply, inflation can look sticky in one part of the economy and structurally disinflationary in another.
That does not mean the Fed should ignore inflation. It means inflation, labor, productivity, and potential growth can no longer be interpreted through a single linear model. Warsh’s point is not that AI automatically means lower rates. His point is that AI changes the reaction function because it changes the economy the reaction function is reacting to.
“We’re no longer going to have to rely solely on data that we get from government agencies with mismeasurement problems that have surveys that are no longer relevant.”
That is also why Warsh keeps returning to data. AI will not only change the economy; it will change how central banks measure the economy. At the hearing, he talked about better inflation measurement, new data sources, and even surveying “a billion prices” to understand the true underlying inflation rate. At the ECB Forum, he expanded that into a broader vision, saying central banks should use new technologies to understand the real economy in a “contemporaneous real-time way.”
He added that the Fed should no longer have to rely solely on government data with “mismeasurement problems” and surveys that are “no longer relevant.” This is a direct challenge to the Fed-watching model. The next edge will not come from obsessing over one adjective in a press conference. It will come from understanding whether the supply side is being transformed faster than legacy data can detect.
“Many of these indicators are echoes of history. We need indicators that tell us what things are.”
The market already understands this lesson in other sectors. Software has been repriced because investors are looking forward into disruption, not backward into reported history. They are asking which businesses are vulnerable to agents, which workflows will be automated, which pricing models will break, and which incumbents will adapt. Fed watchers need the same exercise.
The question is no longer simply whether Warsh sounded tough or patient. The question is whether the Fed’s inherited academic framework can survive a world where digital labor, real-time data, AI-driven productivity, and capex-driven bottlenecks are all arriving at once. The economy is becoming more real-time, more reflexive, and harder for legacy surveys to capture.
“Sometimes unlearning is harder than learning.”
The art of unlearning the Fed does not mean ignoring the Fed. It means listening differently. It requires recognizing that the words hawkish and dovish were built for a world before AI agents. It requires understanding that a central bank can be committed to price stability while also questioning whether the economy’s speed limit has changed. It requires accepting that academic models built from the historical record may be least reliable when the economy is entering a period with no historical precedent.
That was the point of last August’s “Academic Fed” paper, and it is even more urgent now. The Fed cannot rely on an old academic framework for a new AI economy, and Fed watchers cannot rely on one either. Investors who listen to commentary built for the previous regime risk missing the one that is forming now. Warsh’s message is not that central banking should abandon discipline. It is that discipline has to be applied to a different economy. AI is changing the production function, the labor market, the data, and the speed of adjustment. If the Fed has to unlearn parts of its old framework, so do we.
