The AI Washing Test

Block cut 4,000 jobs in January and Jack Dorsey called it an “intelligence tools” decision. The framing was precise: these roles were being replaced by AI capabilities, not eliminated for cost reasons. The distinction matters because it changes how you interpret everything that follows. If AI displaced those workers, the correct response is to retrain, reskill, adapt to a new technological reality. If a post-acquisition company consolidated redundant departments and used AI as the press release’s vocabulary, the correct response is rather different.

The problem is that right now, almost nobody can tell which story is true. And that inability to distinguish real displacement from performed displacement is becoming a structural problem of its own.

The 60/9 gap

A McKinsey survey from late 2025 found that 60% of companies described themselves as “actively transforming operations through AI.” A separate analysis by Epoch AI, using actual deployment data rather than self-reporting, put the number of companies with measurable AI integration into core workflows at roughly 9%. The gap between those two numbers contains an enormous amount of noise, and inside that noise, real people are making real career decisions based on a signal that may be almost entirely fictional.

The 60% figure captures something real, but what it captures is aspiration and positioning, not operational change. Companies say they are transforming because transformation is the expected posture. Investors reward it. Boards demand it. The language of AI adoption has become the language of corporate competence, and any company that doesn’t speak it sounds like it’s standing still. The result is that “AI-driven restructuring” has become available as a framing for virtually any organisational change, whether or not AI is the actual mechanism.

The 9% figure tells a different story. Genuine AI integration into core workflows requires infrastructure investment, data pipeline maturity, workflow redesign, and sustained operational commitment that most organisations have not made. The companies that have done this work look different from those that have not, in ways that are visible if you know where to look but invisible from the outside if all you have is an earnings call transcript.

Why the distinction matters practically

Three constituencies are making decisions based on information that may be systematically unreliable.

Workers facing displacement need to know whether to invest in AI-adjacent skills or whether they were simply caught in a restructuring cycle that would have happened regardless of technology. If your role was eliminated because an AI agent now handles your function, learning to work with that agent, or to build the workflows around it, is a rational career move. If your role was eliminated because the company merged two teams after an acquisition and called it optimisation, the AI reskilling path is a distraction from the actual problem, which is an ordinary labour market transition.

Investors pricing companies on their AI capabilities need to distinguish between companies that have rebuilt their operational architecture around AI and companies that have adopted the vocabulary without the infrastructure. The current information environment makes this distinction nearly impossible from the outside. Earnings calls are optimised for narrative, not accuracy. And the analysts covering these companies are often no better equipped to tell the difference than the general public.

Policymakers drafting regulation and workforce transition programmes need to know the actual scale of AI displacement to calibrate their response. If 60% of companies are restructuring around AI, the policy implications are enormous and urgent. If 9% are, the response needs to be targeted rather than sweeping.

What a real AI washing test would look like

The markers that separate genuine AI-driven change from narrative AI adoption are not mysterious. They are just rarely applied systematically.

Genuine AI integration shows up in capital expenditure patterns. Companies that are actually rebuilding workflows around AI spend measurably on compute, data infrastructure, and integration work. This spending shows up in capex and cloud services line items, not just in the “strategic initiatives” section of the annual report. A company that announces AI-driven workforce reduction without corresponding infrastructure investment is, at minimum, telling an incomplete story.

It shows up in workflow architecture. When AI actually replaces a function, the surrounding workflows change. Reporting structures shift. Quality assurance processes are redesigned. New roles emerge to manage the AI-augmented process. When AI is the framing rather than the mechanism, the surrounding architecture stays the same. The same managers, the same processes, the same QA, just fewer people doing the work.

It shows up in hiring patterns. Companies undergoing genuine AI transformation hire differently: more infrastructure engineers, more prompt engineers and AI workflow designers, more people who sit at the interface between the AI capability and the business process. Companies using AI as vocabulary hire the same profiles they always did, sometimes fewer of them.

None of these markers are individually definitive. But taken together, they form a reasonable diagnostic. A company that announces AI-driven layoffs while increasing AI infrastructure spending, redesigning adjacent workflows, and hiring AI-native roles is probably telling something close to the truth. A company that announces AI-driven layoffs while maintaining the same capex profile, the same organisational structure, and the same hiring patterns is using AI as language, not as technology.

The convenient fiction

The uncomfortable possibility is that AI washing serves both sides of the displacement conversation. Companies get a modernisation narrative that positions layoffs as forward-looking investment rather than cost-cutting. Displaced workers get a more dignified story: you weren’t made redundant because someone cheaper was available, you were displaced by the most powerful technology of the century. There is a strange comfort in being disrupted by the future rather than discarded by the present.

The danger is that this convenient fiction delays the adaptation that actual AI displacement requires. If everyone believes the transformation is already underway, the urgency to prepare for it diminishes. And when the real displacement arrives, as it will for some roles and some industries, the workforce will have spent its preparation time responding to a false signal rather than the real one.

The AI washing test is not about debunking corporate claims for the sake of scepticism. It is about calibrating the signal so that the people who need to adapt can tell whether the adaptation they need is technological or simply economic. Right now, the 60/9 gap makes that calibration nearly impossible. And the longer it persists, the more damage the noise does to the people trying to find the signal inside it.