Challenger, Gray and Christmas reported this month that the tech industry has shed 123,000 jobs so far this year, with AI now surpassing market and economic conditions as the most cited reason employers give for cuts. The number has circulated widely as evidence of displacement. I want to argue that the 123,000 figure, while real, is analytically misleading - not because displacement is not happening, but because the framing directs attention toward the wrong structural problem.

What Employers Are Actually Saying

When a firm cites AI as the reason for a layoff, it is making a claim about substitutability: that a human role can be replaced by an automated process. This claim is sometimes accurate and sometimes a convenient justification for cost-cutting that would have happened anyway. But even when the claim is accurate, it describes only one half of the coordination problem. The more interesting half is what happens to the workers who remain, and whether those workers possess the structural understanding needed to operate effectively inside algorithmically-mediated work environments.

This is where the Challenger data, taken alone, produces a distorted picture. It counts exits. It does not count the much larger population of workers who have not been laid off but who are now expected to coordinate their work through AI systems they do not structurally understand. That population is the organizational theory problem worth examining.

The Awareness-Capability Gap at Scale

Research on algorithmic literacy establishes a persistent gap between awareness and capability. Workers in platform environments develop impressions of how algorithms function - what Gragain, Naab, and Grub (2024) distinguish as folk theories rather than structural schemas - but this awareness does not produce better outcomes. Knowing that an algorithm exists, and even having a rough sense of what it rewards, is categorically different from possessing the schema needed to respond adaptively when the algorithm changes or when the task context shifts.

The 123,000 layoffs represent firms that made a substitution decision. But the firms retaining workers while deploying AI tools are now running an implicit experiment: can workers coordinate effectively through systems they have not been trained to understand structurally? Kellogg, Valentine, and Christin (2020) documented how algorithmic management creates asymmetric information environments where workers bear the adaptive burden while the logic governing their evaluation remains opaque. That asymmetry does not disappear when the firm avoids layoffs. It intensifies.

Routine Expertise Will Not Transfer

Hatano and Inagaki (1986) drew a distinction between routine expertise and adaptive expertise that is directly relevant here. Routine expertise - the procedural mastery of a known task - degrades when the task context changes. Adaptive expertise - the internalization of the principles underlying a task - enables performance across novel variants. Most enterprise AI deployment involves frequent model updates, shifting capability boundaries, and changes to what the system can and cannot reliably do. This is a context that requires adaptive expertise by definition.

The organizational response to AI-driven layoffs has largely been to train retained workers on procedures: how to write a prompt, how to use a specific tool, how to follow a workflow that incorporates a particular model. This is procedural documentation, not schema induction. It produces workers who can execute a known sequence but who will fail when the sequence no longer applies - which, in AI tooling, is a routine occurrence rather than an edge case.

The Number That Matters

Microsoft's AI chief stated this week that Anthropic's models are too expensive and that Microsoft is working to build cheaper in-house alternatives. This is not primarily a cost story. It is a signal that the model landscape firms are training workers to use is unstable. Organizations that have invested in procedural training tied to specific external models are accumulating a competence liability. When the underlying system changes, procedurally trained workers must be retrained from scratch. Workers with structural schemas - who understand why a model behaves as it does, not just how to operate it today - can adapt.

The 123,000 figure will keep rising. That is worth tracking. But the more consequential organizational metric is one that no quarterly report currently measures: the proportion of retained workers who possess transferable structural understanding of the AI systems now mediating their work. Rahman (2021) described algorithmic control as an invisible cage precisely because workers inside it lack the structural knowledge to recognize the constraints shaping their performance. Counting layoffs tells us where the cage doors have closed. It does not tell us whether anyone inside can navigate.

References

Gragain, A., Naab, T. K., and Grub, J. (2024). Algorithmic media use and algorithm literacy. New Media and Society.

Hatano, G., and Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H. Azuma, and K. Hakuta (Eds.), Child development and education in Japan. Freeman.

Kellogg, K. C., Valentine, M. A., and Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366-410.

Rahman, H. A. (2021). The invisible cage: Workers' reactivity to opaque algorithmic evaluations. Administrative Science Quarterly, 66(4), 945-988.