McKinsey's AI client measures comfort with theater, not aptitude

McKinsey is asking some entry-level candidates to have a 10-minute conversation with an AI that simulates a client, according to recent reporting. The company claims the system reflects 80% of scenarios consultants actually face, and humans will evaluate the interaction afterward. It sounds like progress — algorithmic objectivity replacing human bias. It is not. The consulting case interview has never actually predicted job performance — it has only predicted who learned the form.

Case interviews became the industry's primary filter in the 1980s because they appeared meritocratic: test the reasoning, not the resume. What they actually did was screen for candidates who had money for prep courses, family connections inside consulting networks, or both. The form looked like pure logic. It was cultural literacy wearing a neutral mask.

McKinsey's AI system reproduces this entire mechanism at scale by claiming to measure client aptitude when it measures comfort with a specific, trainable interaction choreography. The 80% coverage claim matters less than it sounds. Coverage means the simulation touches eighty percent of client types. It says nothing about whether performing well in a ten-minute AI conversation predicts real consulting work. Only that the scenarios exist. A candidate who has done consulting network events, attended mock case workshops, or used prep software already knows the choreography. Someone equally intelligent but without access to that coaching pipeline has to learn it under time pressure, in the simulation, while being scored.

The version of meritocracy that emerges is narrower than it was before. Not because the AI is biased. Algorithmic systems handle certain types of fairness well. But because the simulation creates a specific, repeatable performance to train for. Means access to coaching becomes the invisible prerequisite. The consulting pipeline was already gated by geography, network, and family connection. An AI client that claims to measure aptitude while actually measuring familiarity with a learnable form just makes the gate look like arithmetic.

When you sit across from a real client, they are unrehearsed and contradictory and bored with your frameworks. The simulation cannot capture that because no AI can simulate genuine human indifference to your intelligence. What McKinsey has built is a more scalable version of what the case interview has always done: filter for people who learned to perform consulting before they had to actually do it. That is not meritocracy. That is efficiency, which is different and worse. It takes a gate that at least operated partially on judgment calls and turns it into a choreography that anyone with enough prep time can master.