After repeatedly telling that some entire job categories will be 'totally, totally gone' due to AI ...

TEXT START: OpenAI CEO Sam Altman is pushing back against the growing belief that AI is directly causing widespread job losses.

The Dissection

Altman is performing emergency damage control. In private testimony to Fed regulators, he delivered the honest assessment: "some areas will be just totally, totally gone." Publicly, he now performs the reassuring pantomime. The article itself exposes the contradiction—the U-turn is the story.

The Core Fallacy

Temporal parochialism: Altman is measuring the current transition phase and mistaking it for the endpoint. His argument rests on what AI doesn't do well now—"long-term, complex task supervision." This is not a structural limit. It is a lag. The trajectory is already mapped: capabilities improve, lag shrinks.

He also commits selection bias dressed as observation. His "companies adopting AI most are hiring most" is measured from his own customer base. These are companies at the frontier of adoption—still in the augmentation phase, still discovering what to do with new capabilities. He is not measuring the companies that have already completed the transition, because that category is growing faster than anyone tracks.

Hidden Assumptions

1. Current capability gaps are permanent features. "AI doesn't do complex supervision well" is treated as a durable constraint rather than a target for investment. Altman has every incentive to understate progress; his investors need the lag to sound long.

2. Human preference for human interaction is a stable variable. Altman assumes this preference is exogenous—fixed by taste rather than shaped by cost and availability. But when AI-generated alternatives are cheaper, faster, and more personalized, preference structures shift. "People want to talk to a person" is a luxury preference. When budgets compress, it becomes negotiable.

3. Aggregate employment is the sum of sector-level observations. Even if some sectors add workers alongside AI, the sectors shedding workers can be larger, faster, and more concentrated in employment share. Adding 50,000 AI-augmented roles while eliminating 500,000 positions is not a net positive.

Social Function

Corporate damage control theater and transition management propaganda. Altman is performing the standard pivot: frighten regulators with accurate analysis to motivate regulatory engagement, then soothe public markets and politicians with optimistic noise. The "people prefer humans" argument is ideological anesthesia for the masses. The sophisticated audience (regulators, enterprise buyers) gets the real data. Everyone else gets the lullaby.

The Verdict

Altman is not retreating from the Discontinuity Thesis. He is performing the necessary social management that precedes it. You cannot tell the public "jobs are totally gone" and maintain the political legitimacy required to capture the transition. So you deliver the scary truth to power and the soothing lie to everyone else.

The gap between his Fed testimony and his CNBC performance is not a contradiction. It is the correct strategy: manage the collapse, don't announce it.

The Discontinuity Thesis framework remains intact. Current AI adoption patterns represent the lag phase—augmentation before replacement, collaboration before elimination. The jobs being added now are transition roles, not structural endpoints. As capabilities improve, those supervisory and collaborative positions become the next automation targets.

The question is not whether Altman believes this. The question is whether he needs you to believe it. He does.