While financial markets are laser-focused at the moment on geopolitical risks and the near-term inflationary consequences of the spike in energy prices, beneath the surface, a growing consensus is forming that core US inflation dynamics are shifting back toward their pre-COVID configuration. But if we take that baseline seriously—and overlay the rapidly rising probability of meaningful labor market disruption from generative and agentic AI—the conclusion may be uncomfortable for the inflation market. Markets should arguably be pricing a negative core inflation risk premium beyond the near term, not a positive one, which would be a flip from our thinking entering this year and underscores how quickly dynamics are changing.

Before turning to AI, consider the underlying arithmetic of inflation. Before the COVID-19 pandemic, the US core CPI ecosystem was remarkably stable. Core goods inflation oscillated around 0%. Rent inflation ran roughly 3.0–3.5%. Core services ex-rents sat between the two in a way that delivered a weighted average of about 2%. This distribution held for years and provided the anchor for policy and market pricing.

Now, if we assume tariff pass-through fades and core goods return to something like 0% in H2 2026, the debate shifts squarely to the other two components. Rent inflation is already widely expected to run below its pre-COVID trajectory. Incoming data from new lease measures, construction pipelines and vacancy rates all point in the same direction. If rents are no longer delivering 3.0–3.5% and core goods inflation returns to flat, then simple math requires core services ex-rents to run higher than they did pre-COVID just to keep the core basket near 2%.

This part of the inflation debate has not received adequate attention, in our view. The bar for supercore inflation is now higher. Yet, wage tracking—a primary input into supercore—has already decelerated to its pre-COVID pace. The Atlanta Fed wage tracker and Barclays’ state-space decomposition show wage momentum sitting at or slightly below where it would need to be to generate 2% inflation in this new regime, given softer rents and flat goods.

This raises the uncomfortable question: if wage growth is already too soft for the new core inflation arithmetic, what happens if productivity-enhancing or labor-substituting technologies accelerate?

That brings us to generative and agentic AI.

Over the past six months, the step-function improvement in model capability—and the plausible emergence of agentic architectures that can autonomously complete multi-step tasks—has shifted the probability distribution of labor-market outcomes. The viral blog post last week resonated precisely because it articulated a widely felt, but poorly formalized, intuition, namely that we may be approaching the point at which AI substitutes not just tasks, but workflows. The reaction across industries speaks for itself. Block’s announcement of significant layoffs—explicitly tied to automation—gave that intuition corporate form.

Dismissing these developments on the grounds that there is no evidence in the labor data misses the underlying dynamics. That argument is analogous to saying hurricane forecasts are wrong because there is no current flooding. By the time the labor market shows stress, the repricing will already have occurred. Markets do not wait for realized outcomes when the distribution itself has shifted.

Another critique—that the tools are not fully there—also feels increasingly hollow. The metaphor is not the slow emergence of a mature technology; it is the comedic moment in The Office when Michael and Dwight follow the GPS into a lake. The software was not perfect, but the system still redirected behavior in ways that had real consequences. A sober assessment of where AI systems will be in the next several years, based on observed improvement curves and announced roadmaps, suggests sharply rising probability of meaningful disruption, which could disproportionately weigh on wage momentum.

If the market begins to internalize the possibility that AI could soften labor demand before it meaningfully lifts productivity, then the risk to inflation is skewed decisively lower. And the current pricing of inflation compensation does not fully reflect these downside risks. Moreover, at a certain point, the evolving AI shock should become a positive productivity shock, which should be disinflationary, at least in the medium term; this point is more compelling than the more dire negative aggregate demand scenario laid out in some of the more alarming prognostications.

And yet, 2yfwd3y CPI swaps remain above 2.3%, while 2yfwd3y real rate swaps only briefly dipped below 1% and are still 20bp above their lows from last September. If investors begin assigning greater weight to the AI-disinflation scenario—whereby wage growth undershoots just as the supercore hurdle has risen—the reaction function across rates markets could shift quickly. A downward repricing of inflation risk premia would likely push the front end of the nominal yield curve into a rally, compress real yields, re-steepen the curve, and break the current low-volatility equilibrium that has defined markets in recent months (though that is already at risk, given the conflict in Iran). The near-term risks are more two-way, as exemplified by the energy price shock that drove front-end inflation higher on Monday on a spike in geopolitical tensions in the Middle East, to say nothing of massive AI capex that could be supportive for inflation pressure, or broadly rising electricity prices. That 1y1y CPI swaps actually fell slightly on Monday could indicate how much of a read-through there may end up being from the Iran situation to underlying inflation momentum.

The question is not whether AI has already disrupted the labor market; it is whether the probability of such disruption is rising fast enough to influence the distribution of future inflation outcomes. On that dimension, the evidence is increasingly difficult to ignore.

Oh, and as a final point… this piece was largely drafted by AI, executing an extensive prompt laying out the arguments we wanted to make and examples we wanted to use in one iteration, with only slight tweaks and edits required afterwards. While analysts will remain the source of ideation and need to write their thoughts in the form of a comprehensive prompt, AI is enabling better and faster execution.