Why the New Fed Chair Talks Like a Startup Founder

Opening

On July 14, newly appointed Fed Chair Kevin Warsh made a statement at his first congressional hearing since taking office: “AI is probably the most consequential change to our economy since I’ve been an adult.” He paired it with a promise to consign the “tax” of inflation to the past.

If that gives you déjà vu, your instincts are right. I’ve already covered this story twice in this newsletter — investment measured in the ₩1,000 trillion range against dead-quiet economic indicators. In We Poured In $250 Billion. Why Doesn’t GDP Show It?, I laid out the lag inherent to general-purpose technologies; in 47 Years Ago, the Office Automation Era, I dug into the gap behind “it feels like 3x, but the data says 1.8%.” The $250 billion figure from that earlier piece has, per the FT’s tally, since grown to a cumulative $725 billion.

So today I won’t repeat the same question. I’ll flip it. Instead of “where is the productivity,” I’m asking “what is a productivity gain that hasn’t even arrived yet already buying?” Here’s the conclusion up front: AI productivity is circulating as a policy narrative before it shows up as an economic indicator. And the first place that narrative is being cashed in is U.S. monetary policy.

---

Three Pieces of Evidence Against “Wait and It Will Come”

The conclusion of my previous two pieces was a kind of optimistic deferral: general-purpose technologies inherently take time, so let’s wait until organizations redesign how they work. But over the past month, evidence unfavorable to that “wait and it will come” hypothesis has emerged from three directions.

First, the depth of adoption has stalled. According to the Real-time Population Survey (RPS) run by St. Louis Fed economists, 45% of the U.S. working-age population used generative AI at work at least once a week as of Q2 2026. But the share using it “daily” has been stuck sideways at around 1 in 10 for over 1 year. Usage time among users runs about 30 minutes a day, and the time savings calculated from that come to roughly 10 minutes a day. The lag hypothesis rests on the premise that adoption keeps deepening — but while breadth has grown, depth has flatlined. A tool you use occasionally just substitutes for search; only a tool you use daily changes how you actually work.

Second, even the productivity gains that have already shown up may not belong to AI. Chair Warsh cites the high-2% range growth in structural productivity¹ over the last 4 quarters as grounds for optimism. But when Barclays economists adjust for cyclical capacity utilization, that figure drops to the low-1% range. Their explanation: during the post-pandemic labor-hoarding² period (2022–24), when companies kept staff on even as demand cooled, measured productivity understated true capability; now that those workers are being let go, the statistics are mechanically bouncing back. It’s a cyclical reversion, not a technological leap.

Third — and this is the part that stings most — evidence has begun to move past “we don’t see it yet” toward the more active claim that “even the fast adopters look no different.” If AI were truly lifting productivity, it should show up first in early-adopting industries like finance, IT, and consulting. But when Barclays ran the correlation between industry-level adoption rates and productivity gains through multiple specifications, the result was “statistically indistinguishable from 0.” A July 17 analytical note from Fed economists paints the same picture: grouping industries into high, medium, and low AI-adoption tiers, the productivity trends of all three groups tracked each other closely over time.

None of this proves the lag hypothesis wrong. But this is the first time the shelf life of the word “yet” has come under real data-driven interrogation.

---

The Bill Is Already Being Paid, Before the Productivity Has Arrived

Yet regardless of this unresolved status, that productivity is already being spent. And the place it’s being spent is monetary policy.

Let me lay Chair Warsh’s statements out chronologically. Before taking office, in a Wall Street Journal op-ed last November, he wrote that “AI is likely to be a substantial disinflationary force” and that “a 1-percentage-point rise in annual productivity growth doubles living standards within a generation.” Earlier this month, at a European Central Bank forum, he cited high-2% range structural productivity as grounds for “reason to be optimistic.” And at this hearing, he said AI appears to be raising productivity without displacing workers — tying the promise of ending inflation and the direction of rate cuts into a single sentence.

The logic is simple: if productivity rises, growth can continue without price pressure, which makes it safe to cut rates. If productivity has genuinely risen, this is textbook-correct. The problem, as we just saw, is that the premise isn’t in the data yet.

The FT traces the lineage of this narrative in an interesting way: a magic force that grows the economy without stirring workers’ bargaining power. Sound familiar? It’s quantitative easing³. The one difference is that QE showed up as trillion-dollar figures on the Fed’s balance sheet, while the AI-productivity narrative sits off the books entirely — a stimulus with no visible cost. Between a White House that has publicly wanted rate cuts and a new chair who took office promising a policy “regime change,” there’s no more convenient bridge than this.

There’s a contrast worth watching here — a temperature gap inside the very same building. The July 17 note from Fed staff economists carefully hedges, saying “no difference is visible in productivity trends across industries,” while the institution’s own chair is stating policy direction premised on optimism at a congressional hearing. The research wing’s tense is “not yet”; the policy statements’ tense is already “already.”

Another feature of this narrative is that, by design, it’s hard to falsify. The sentence “it’s not in the statistics yet, but it’s coming” survives no matter what data comes out. The FT notes that media citations of Solow’s productivity paradox are on pace to break their record for the 2nd year running, and I read that number as a demand indicator for the word “yet.” If a 40-year-old sentence is selling well again, that tells you how many people currently need the language of deferral.

---

What Happens When the Narrative Outruns the Data

Start with the question of sequence. There’s a difference between cutting rates after confirming a productivity increase, and cutting rates by assuming one. The former is policy; the latter is a bet. If the assumption is right, you’re a hero; if it’s wrong, the bill comes due as inflation. That’s why Barclays deliberately inserted, in its report’s conclusion, the line: “we oppose easing monetary policy on this basis.”

There’s also a maturity mismatch. If AI really is, as Chair Warsh says, “the most consequential change,” history shows such transitions play out over decades — and the gains have typically flowed not to incumbents but to newcomers. That’s how manufacturing automation went, and how online retail went. Meanwhile, a Fed chair’s term is 4 years. Front-load a decades-long transition’s payoff onto a 4-year policy clock, and someone, somewhere, has to make up the difference. And costs pile up even while the narrative holds: asset prices rise on the strength of that optimism, data centers get built, hiring and budgets get reallocated. On the day the narrative gets corrected, what adjusts isn’t a sentence — it’s these real, physical things.

And this isn’t just an American story. Sentences that begin “since AI will raise productivity” are proliferating rapidly in Korean policy statements and corporate announcements too. The sentence itself is innocent. What matters is the decision that follows it — what budget, what reorganization, what interest rate and what investment that narrative is pushing through. And who ends up holding the bill if the assumption turns out wrong.

---

Oz’s Lens

Honestly, this rhetoric wasn’t unfamiliar to me at all. It’s the language of startups.

In go-to-market strategy consulting, I’ve run into the sentence “the metrics aren’t there yet, but they’re coming” more times than I can count. A funding round without traction gets raised on narrative, and that sentence is the standard grammar of that narrative. What’s striking is hearing that same grammar now coming out of the institution that, of all institutions in the world, should be the most careful. When a startup loses that bet, the loss lands on investors. When a central bank loses the same bet, the bill gets issued to the entire population, in the form of inflation.

One more thing from my data background: the stronger a narrative gets, the more numbers tend to be measured to fit it. When the U.S. Census Bureau survey broadened the wording of its question on AI use, the index jumped nearly 10 points. A domestic Korean survey that sampled companies’ AI/IT staff specifically produced a 55.7% adoption rate. Neither the 21% from the U.S. survey, representative of all businesses, nor the 55.7% from the Korean survey of designated staff, is a lie. But which number gets put center stage tells you exactly whose narrative currently needs what.

In the interest of balance, let me note: not showing up in the statistics isn’t proof of no effect. I’ve seen the pattern countless times in tech markets, where the forecast is right but the timing and path are wrong, and AI productivity will eventually arrive too. Today’s point isn’t that it won’t come. It’s the fact that a central bank is, right now, betting that it will come within this rate-cutting cycle.

---

Closing

Let me sum up today’s piece in three lines. Against “wait and AI productivity will come,” unfavorable evidence — stalled adoption, cyclical-capacity noise, a zero correlation — has begun piling up for the first time. Meanwhile, that unconfirmed productivity has already started circulating in the form of grounds for rate cuts. So the thing to watch now isn’t the productivity statistics themselves — it’s who is cashing in what, using that narrative.

Next time you hear an “AI productivity” statement, check just two things. What is the speaker currently justifying with that narrative? And is there a falsification condition — a “what data would make me drop this forecast”? A forecast with no falsification condition isn’t analysis. It’s sales.

Has there been a moment in your own organization when “the numbers aren’t in yet, but they’re coming” won out over actual measurement? Tell me in the comments what decision that narrative pushed through, and how it turned out. If enough examples come in, I’ll devote a future issue to “moments when narrative beats data.”

---

💬 If you’ve witnessed a moment when the “it’s coming soon” narrative beat out actual data, tell me in the comments. I’ll factor it into a future issue. 📨 If you have a colleague who has to make a lot of judgment calls between AI investment and policy statements, share this piece with them.

---

References & Further Reading

Primary sources

• FT Alphaville, “Is AI productivity growth in the room with us right now?”, Financial Times, 2026.7. Link ··· This is where today’s piece starts. The key charts and arguments from Barclays’ non-public report are laid out here.
• Soto, P., Thieu, M. & Allen, J., “The AI Buildout and the Economy: Publicly Available Data to Assess AI’s Impact”, FEDS Notes, Federal Reserve Board, 2026.7.17. Link ··· The comparison of productivity trends across high/medium/low AI-adoption industries is the core of this note. Since it uses only public data, you can reproduce it yourself.
• Bick, A., Blandin, A. & Deming, D., “The Rapid Adoption of Generative AI”, NBER Working Paper 32966, 2024 (updated quarterly). Link ··· The methodology paper for the Real-time Population Survey (RPS), the source of the adoption-rate and usage-time figures.
• Warsh, K., “The Federal Reserve’s Broken Leadership”, The Wall Street Journal, 2025.11. Link ··· The clearest statement of Warsh’s pre-office thinking. The “disinflationary force” quote comes from here.
• FedScoop, “Fed chair says AI ‘hasn’t displaced workers’ so far, has boosted productivity”, 2026.7. Link ··· A summary of the July 14 House hearing remarks.
• U.S. Census Bureau, “Business Trends and Outlook Survey”, 2026. Link ··· The source for the 21% business-adoption rate and the story of the changed question wording.
• CIO Korea, “85% of Korean Companies Have Adopted Generative AI in 2026, 8 in 10 Expanding Budgets”, 2026. Link ··· A survey of 749 respondents in Korea by Megazone Cloud and Foundry. Recommended for comparing measurement design against the U.S. statistics.

Background

• “Productivity paradox”, Wikipedia. Link ··· Lays out the 40-year debate around Solow’s paradox. Seeing how it was resolved in the 1990s changes how you feel the weight of the word “yet.”

Past issues worth reading alongside this one

• We Poured In $250 Billion. Why Doesn’t GDP Show It? ··· This is part one of today’s piece. I built the lag hypothesis for general-purpose technologies here.
• History Repeats? 47 Years Ago, the 📠 Office Automation Era ··· A piece diagnosing the “feels like 3x, measures like 1.8%” gap as an organizational problem. Worth reading alongside today’s new evidence.

---

📝 Glossary

The author, Kwangseob Ahn, is a professor of business administration at Sejong University and lead consultant at OBF (Oswarld Boutique Consulting Firm). He teaches statistics and data analysis, including business data management and business analytics, while leading GTM and AI strategy consulting in the field, designing the seam between technology and business. He has published academic research on a memory architecture for AI dialogue systems (HEMA) and runs Daily Arxiv, a daily curation of global AI papers. He holds a master's from Korea University's Graduate School of Technology Management and a KMBA. He is the author of Homo Brainless: The People Who Outsource Their Thinking.

Footnotes

1. Structural productivity: The productivity growth rate of the economy’s underlying fundamentals, with temporary swings from booms and busts stripped out. It’s used as an input for gauging potential growth. ↩

2. Labor hoarding: The practice of companies keeping staff on instead of laying them off even as demand falls. It happens when rehiring is difficult or hiring costs are high; during such periods, output falls with the same headcount, so measured productivity looks lower. ↩

3. Quantitative easing (QE): A policy in which a central bank buys large quantities of assets like government bonds to inject money into the market. It’s used as a stimulus tool when rates can’t be cut further, and its scale shows up directly on the central bank’s balance sheet. ↩