I see the disconfirming cohort at primary: one engineering-labor firm grew revenue 15% while net income fell 17%; another's quarterly revenue declined 4.7% with AI immaterial; a third pairs "2,500+ generative-AI projects" rhetoric with low-single-digit growth guidance. The tools creating demand compress billable hours. The research self-corrected: its claim one firm cut guidance was falsified — it raised guidance — and the correction is carried openly. MIT finds ~95% of enterprise AI pilots show no measurable P&L impact; vendor-built deployments succeed about twice as often as internal ones.