The intuitive picture of AI job displacement goes something like this: low-wage, routine workers in warehouses and call centres are most at risk, while high-earners in creative and knowledge-intensive roles are relatively insulated. New data is complicating that picture in interesting ways — and a warning from a Federal Reserve governor that monetary policy cannot fix AI-driven unemployment adds a political dimension that is only beginning to register.
CNBC reports that top earners are now more anxious about their employment than lower-income workers, reversing the expected pattern. The logic, on reflection, is not hard to follow. Lower-wage service jobs involve physical presence, interpersonal interaction, and the kind of contextual judgment in messy environments that AI still struggles with. Higher-income knowledge work — legal analysis, financial modelling, software development, content creation — is precisely where large language models have made the most rapid and measurable gains. The people whose jobs involve sitting at a computer and producing documents or analysis are, it turns out, the people most exposed to tools that sit at computers and produce documents and analysis.
The Washington Post's interactive jobs-at-risk map, updated for 2026, reinforces this: roles at the 80th income percentile and above show the highest AI automation exposure scores, not the lowest. That does not mean those jobs will disappear — automation exposure has historically translated into task transformation more than wholesale job elimination — but it does mean the anxiety is rational, not just noise.
What makes this particularly awkward for policymakers is the remark from Federal Reserve Governor Lisa Cook, reported by Bloomberg, that the Fed may simply not have the tools to counteract AI-driven job loss. Traditional monetary policy addresses demand shortfalls and credit conditions; it is poorly suited to a supply-side productivity shock that eliminates specific categories of work regardless of interest rates or liquidity. If AI compresses the value of certain labour, cutting rates does not bring that value back. The observation is not new in academic circles, but having a sitting Fed governor say it out loud is notable.
The policy vacuum this creates is real. The US government has so far leaned heavily on the "AI creates more jobs than it destroys" argument, citing historical precedent with previous technology waves. That argument may well be correct in aggregate over a long enough time horizon — but it offers little comfort to a mid-career financial analyst whose core tasks are now cheaper to automate than to staff, and it says nothing about the distributional question of who captures the productivity gains.
Anthropic's own labour market research, released earlier this year, found that AI assistance is already measurably shifting what kinds of tasks workers spend time on, with routine cognitive tasks declining and oversight, communication, and judgment tasks growing. Whether that is a story about augmentation or displacement depends on whether the workers whose routines are being automated can successfully pivot to the growing tasks — a question the data does not yet answer cleanly.
What is becoming clear is that the "tsunami" metaphor that keeps appearing in analyst commentary is apt in one specific way: the water receding before a wave looks a lot like calm. Job losses from AI have not yet shown up dramatically in aggregate unemployment statistics, and for many knowledge workers, AI tools have so far felt like productivity boosters rather than existential threats. The question keeping economists and governors up at night is whether the current phase is the calm, and what comes after it.