The Announcement

This week, OpenAI, Anthropic, Microsoft, and Amazon jointly backed a new nonprofit organization led by former Commerce Secretary Gina Raimondo, with an explicit mandate to prepare workers for AI-driven job displacement. The organization is notable not just for its funding constellation but for what it reveals about how leading AI firms conceptualize the workforce problem they are helping to create. The framing, predictably, centers on training. Workers need new skills. Organizations need to adapt. The nonprofit will presumably deliver content, credentials, or curriculum to bridge the gap.

This is the wrong model, and the organizational theory literature gives us precise language for why.

What the Nonprofit Gets Wrong About Competence Development

The implicit logic behind workforce AI initiatives of this type is what I would call the competence-transfer assumption: that skills developed in one context (a training program, a credential course, a workshop led by industry partners) will transfer reliably into the algorithmically-mediated work environments those same sponsors operate. Decades of cognitive science research should make us skeptical of this assumption.

Hatano and Inagaki (1986) drew a foundational distinction between routine expertise and adaptive expertise. Routine expertise produces reliable performance within familiar conditions. Adaptive expertise produces reliable performance across novel and shifting conditions. Platform-mediated work environments, by design, are constantly novel. Algorithms change. Interfaces change. The structural logic of how outputs are rewarded shifts without notice. A training program that teaches workers how a specific AI tool works today is producing routine expertise for an environment that will not remain static long enough for that expertise to matter.

Kellogg, Valentine, and Christin (2020) documented exactly this failure mode in their review of algorithmic management. Workers subject to algorithmic oversight develop detailed folk theories about how systems evaluate their performance, but those folk theories frequently diverge from the actual structural logic of the algorithm. Knowing that an algorithm exists, and even having a working hypothesis about how it behaves, does not produce the adaptive capacity to respond when it changes. This is the awareness-capability gap, and it is the central problem that an initiative like Raimondo's nonprofit will reproduce if it does not address the distinction explicitly.

The Structural Problem with Industry-Funded Workforce Development

There is a second issue that sits beneath the competence question, and it is an organizational one. The same firms funding this nonprofit - OpenAI, Anthropic, Microsoft, Amazon - are the firms whose systems will govern the work environments the trained workers enter. Rahman (2021) described this structure precisely in the context of platform work: the organization that sets the rules of coordination is also the organization that benefits from worker compliance with those rules. The invisible cage is built by the same actors who offer the map.

This is not a conspiracy claim. It is a structural observation. When firms fund workforce development for environments they control, the training curriculum will, consciously or not, reflect the firm's interest in producing workers who are competent within the current architecture rather than workers who can reason about and adapt to architectural change. The distinction is subtle but consequential. The first produces dependency. The second produces genuine platform literacy.

Schor et al. (2020) documented how platform dependence operates through a combination of economic necessity and competence asymmetry. Workers who lack structural understanding of the platforms they depend on are less able to negotiate, adapt, or exit. A nonprofit that delivers procedural AI training, without delivering schema-level understanding of how algorithmic coordination actually functions, risks deepening that asymmetry while appearing to address it.

What a Structurally Sound Alternative Would Look Like

The ALC framework I am developing at Bentley offers a specific prediction here that is relevant beyond my dissertation context. Training that induces structural schemas - accurate representations of the relational logic governing algorithmically-mediated environments - should produce better transfer outcomes than training that delivers platform-specific procedures. Gentner's (1983) structure-mapping theory provides the cognitive mechanism: transfer occurs when learners map structural relations across domains, not when they memorize surface-level procedures.

For a workforce development initiative at the scale Raimondo's nonprofit is proposing, this distinction has real design implications. The question is not whether workers can use a specific AI tool. The question is whether workers understand enough about how algorithmically-mediated environments distribute outcomes to adapt when those environments change. The first question is answerable with a tutorial. The second requires curriculum design grounded in organizational and cognitive theory that the current announcement gives no indication of prioritizing.

Why This Moment Matters for Organizational Theory

The Raimondo initiative is significant precisely because it is large-scale, well-funded, and framed as a solution. Initiatives of this profile set precedents for how workforce AI problems get institutionalized. If the dominant model becomes industry-funded procedural training delivered through nonprofit intermediaries, we will have an answer to the workforce displacement problem that is organizationally legible and theoretically inadequate. The coordination failure will not disappear. It will simply become harder to see.