The Fed’s Hidden Powers and Its AI Future

A hidden power in Section 2 of the FRA, a glimpse of an AI-run FOMC, and the prospect of productivity-driven deflation in the age of AI.

After several decades of studying and writing on the Fed, I thought I knew the institution well and was past the point of being surprised by it. This week, however, I was reminded that there is much about the Fed I do not know. This realization hit me while reading the Federal Reserve Act (FRA). In this newsletter, I will share the revelation I discovered about the Fed, along with two fresh insights into how AI could radically reshape the Fed in the years ahead.

Before we jump into these three new things I learned about the Fed, here is a video clip from this week’s podcast with SEC Commissioner Hester Peirce. We had a great conversation on rethinking surveillance, crypto rules, the SEC’s crypto task force, capital formation, and Treasury clearing. Check out the show!

The Amazing Section 2 of the Federal Reserve Act

CEA Chair Stephen Miran went before the Senate Banking Committee this week for his confirmation hearings as a Fed governor. He did a good job making the case for and showing his commitment to Fed independence, but these words rang hollow to many Senators after he revealed that he plans to take an unpaid leave of absence and keep his CEA job while serving as a Fed governor. The Senators were startled by this admission. They pushed him on how he could keep his independence from the President and maintain a CEA job at the same time. Senator Ruben Gallego’s question captured their concerns well:

"I don't understand why you don't see why so many people are worried that you're going to go into a temporary position at the Fed, hold your other position within the White House, and we don't think the Sword of Damocles is gonna be hanging over your head if you don't lower interest rates?"

There is both an irony and tragedy to Miran’s plans. The irony is that he and his coauthor argued exactly against this very type of behavior in a 2024 policy paper titled Reform the Federal Reserve’s Governance to Deliver Better Monetary Outcomes. This is actually an interesting paper and I interviewed Stephen and his coauthor, Dan Katz, about their proposals in it on the podcast back in 2024. The tragedy is that any of Stephen’s ideas for Fed reform, including those in the paper, will now be viewed with skepticism. Observers, in good faith, can question whether Stephen as a Governor will be championing Fed reforms because he believes in it or because President Trump wants it.

One example of this is Stephen’s proposal to reform the regional Federal Reserve banks. They have a lot of power, lack the democratic accountability that Fed governors face, and manage districts that are increasingly outdated. You might disagree with the details of his specific plan to reform these banks, but it is a good-faith argument laid out in his paper. If he championed reforming the regional Fed banks, would it be viewed as his own idea or as President Trump’s initiative?

I bring up this specific example because Saleha Moshin reports in Bloomberg the following:

The Trump administration is reviewing options for exerting more influence over the Federal Reserve’s 12 regional banks that would potentially extend its reach beyond personnel appointments in Washington, according to people familiar with the matter.

What is being reported is that Trump could allegedly influence the regional banks since the bank presidents have to be reapproved by the Board of Governors (BoG) every five years. If Trump got a majority on the BoG, then he could influence the regional bank presidents. Moshin reports that Trump’s team claims the “goal isn’t to make the central bank more dovish, but to scrutinize how regional presidents are vetted and chosen since they are not Senate-confirmed.”

Okay, so here is where the amazing Section 2 of the Federal Reserve Act (FRA) comes into play. I assumed that BoG ability to influence the regional banks was limited and so folks really were making much ado about nothing here. I decided to go read the section of the FRA that speaks to the formation of the regional banks for myself to confirm my suspicion. What I found, instead, was mind-blowing. Section 2(1) contains the following passage:

The [Federal Reserve] districts thus created may be readjusted and new districts may from time to time be created by the Board of Governors of the Federal Reserve System, not to exceed twelve in all

Wow! The BoG can actually redraw the map for the Federal Reserve districts and banks! This means the BoG could, in theory, eliminate some regional banks and form new ones as long there are at least 8 and no more than 12. Completely revamping the Fed districts appears radical, but anything seems possible under President Trump. Something to watch!

This is not the first time Section 2 of the FRA surprised me. Just a few weeks ago I had Paul Kupiec of the AEI on the podcast to talk about the Fed’s balance sheet. I was blown away to learn from him that the FRA allows the Fed to recapitalize itself by requiring member banks to pay in their full capital subscription. Guess where that passage is found? Yep, Section 2 as well:

shall be required within thirty days after notice from the organization committee, to subscribe to the capital stock of such Federal reserve bank in a sum equal to six per centum of the paid-up capital stock and surplus of such bank, one-sixth of the subscription to be payable on call of the organization committee or of the Board of Governors of the Federal Reserve System, one-sixth within three months and one-sixth within six months thereafter, and the remainder of the subscription, or any part thereof, shall be subject to call when deemed necessary by the Board of Governors of the Federal Reserve System, said payments to be in gold or gold certificates.

To be clear, this would not be enough to fully recapitalize the Fed. Paul also noted that the Fed could stop paying dividends to the banks. Both of these options seem unlikely to happen, but still an amazing and unexpected part of the FRA. That is why I call it the Amazing Section 2 of the FRA.

The Amazing AI Future of the Federal Reserve

I often joke on the podcast and X that one day the Fed will be largely run by AI. Well, my humor might become reality far sooner than I imagined. That is one of the big takeaways I got from a new paper by Sophia Kazinnik and Tara Sinclair.

Their paper is titled FOMC In Silico: A Multi-Agent System for Monetary Policy Decision Modeling. In it, they build a digital Federal Open Market Committee powered by large language models (LLMs). Each FOMC member is represented by an AI “agent” trained on that policymaker’s biography, historical views, and regional economic conditions. The system then ingests real-time macro data, speeches, and financial news before launching into a simulated committee meeting.

What makes the paper innovative is its dual-track design. Alongside the LLM-driven debate, the authors run a rational Bayesian voting model — a purely statistical benchmark without the frictions of persuasion or politics. The comparison is striking: in baseline July 2025 conditions, both tracks converge on a federal funds rate around 4.4%, right where the real FOMC sat. But once political pressure or a negative jobs shock is introduced, the LLM committee starts to dissent, polarize, and drift in ways that look remarkably like the actual FOMC under stress. The result is a powerful demonstration of how AI can replicate not only the “rational core” of central banking, but also the messy, human dynamics that shape real-world policy outcomes.

For me, the implications are hard to ignore. If an LLM-driven committee can so accurately reproduce the deliberations of the Fed, then much of the central bank’s work could be delegated to AI. One could imagine a future where the Fed Chair and Vice Chairs remain in place as the public face of policy, bearing political accountability, and ensuring legitimacy while the rest of the committee is staffed by AI agents trained on the reasoning of strong monetary economists and past central bankers. To paraphrase Ken Jennings, we may soon be saying “I, for one, welcome our AI overlords on the FOMC”

The Amazing Secular Deflation Future of the Federal Reserve

Speaking of an AI future, imagine if Artificial General Intelligence (AGI) truly arrives and ushers in an age of robust productivity growth that dramatically raises real economic growth. How should monetary policy respond to such an environment? Put differently, what would be optimal monetary policy if real GDP growth averaged, say, 6% a year? Would we really want 2% inflation with in that environment?

One could imagine that providing sufficient monetary accommodation to generate 2% inflation in a world of rapid productivity growth could generate some destabilizing asset bubbles. This is a point well articulated by Lawrence Christiano, Cosmin Ilut, Roberto Motto & Massimo Rostagno in their 2010 Jackson Hole Symposium paper.

Imagine instead, the Fed aimed to stabilize nominal income growth at say 4% a year. That would imply a steady-state world of 2% secular deflation with real growth at 6%. Such a scenario seems so foreign to us, but could it work? What would happen to financial intermediation in a world of secular deflation? Would we be stuck at the effective lower bound (ELB)?

The reason we worry about these issues in a deflationary environment is because we typically invoke aggregate demand-driven deflation. In those environments, there is a debt-deflation spiral as nominal incomes fall and nominal interest rates collapse. In theory, it is a very different story in a world with aggregate-supply driven deflation. In this setting, one can still have stable nominal incomes and the ability to service nominal debt even if the price level is mildly falling. Stable nominal income growth and the expectation of real growth together should keep financial intermediation from collapsing. Also, rapid real economic growth raises real interest rates and should offset declines in the expected inflation term on nominal interest rates.

Recent evidence from Bryan Cutsinger and Casey Pender directly addresses these theoretical questions with empirical evidence. They separate supply-driven from demand-driven deflation using a sign-restricted Bayesian panel VAR on annual data for twelve gold-standard economies (1880–1900). They track output, the price level, short-term nominal rates, and proxies for financial intermediation. Their headline result is simple and powerful: when prices fall because of positive aggregate supply shocks, output rises, while nominal rates and financial intermediation do not deteriorate. By contrast, aggregate demand-driven deflation pulls down nominal rates and weakens financial intermediation. Their forecast error decompositions show sizable supply contributions, and the policy takeaway is to “look through” supply-driven deflation rather than offset it.

In my own work on the U.S. postbellum period (1866–1897), I found that on average the price level fell by a little over 2% per year while real GNP grew nearly 4% annually, with real wages advancing as well. What stands out to me is that this long stretch of secular deflation coincided with robust economic growth rather than stagnation. Moreover, nominal interest rates never hit the ELB and financial intermediation actually deepened over the period, even though there were episodes of financial turmoil in the 1870s and 1890s. Additionally, when I ran VARs on this era, I found that nominal rigidities were meaningful—aggregate demand shocks did move output—but the broader trend of a falling price level was still benign because it was supply-driven. For me, this makes the postbellum U.S. experience a powerful historical example of how an economy could thrive under productivity-led deflation.

Looking ahead, the arrival of AGI could recreate the kind of productivity-driven deflation we saw in the postbellum United States and that Cutsinger and Pender identify across a number of countries. Their work, along with my own, shows that such deflation is not something to fear so long as nominal incomes remain stable. That is precisely why an NGDP-targeting framework makes sense in an AGI-driven world of rapid economic growth: it would allow prices to fall naturally with productivity while safeguarding debt contracts, financial intermediation, and macroeconomic stability. In short, if AI agents end up running both the economy and the FOMC, the best way to harness their potential is to let them deliver the gains of productivity through lower prices while anchoring policy on the steady growth of nominal income.

The real test, then, of central banking in the age of AGI will not be whether we can hold the line at 2% inflation, but whether we can reimagine stability around a stable growth path in nominal income.