The article draws a structural parallel between the management consulting industry's "confidence trick" — documented by economists Mariana Mazzucato and Rosie Collington in The Big Con — and today's agentic AI vendors. Both exploit an asymmetry: polished, confident outputs meet clients whose capacity to evaluate them has atrophied through repeated delegation. Consultants package generic analyses as strategic certainty; AI systems, fundamentally probabilistic pattern matchers with no accountability, deliver plausible but often generic or wrong results. The mechanism is "unlearning by not doing": the less an organization performs a function internally, the less it knows how to do it, the more it needs outside help, and the less it builds the knowledge that would make outside help unnecessary.

This dynamic played out in public administration's "steer more, row less" reforms of the 1980s, which backfired for functions where doing and directing cannot be separated. With agentic AI, the forfeiture goes further — agency passes to systems with no real accountability to the user. Early AI wins (drafting, coding, summarizing) create rational incremental delegation, while subsidized pricing — akin to early cloud and SaaS models — builds integration depth that later makes switching costly. Vendors race to capture market share to justify growth-stock valuations, pricing well below true delivery costs.

The risk spans three levels: students bypassing the cognitive struggle that turns writing into understanding; companies shedding tacit institutional knowledge needed to evaluate AI output; courts deferring to opaque algorithmic risk scores. Reclaiming agency requires recognizing that comparative advantage logic turns toxic when it carves out the essential, strategic work that defines an entity's identity.