The Policy That Exposed What Organizations Cannot Measure

A quiet but significant shift is underway in corporate performance management. A growing number of companies are now tying career advancement directly to AI tool adoption, according to recent reporting. Workers who demonstrate measurable AI use are being prioritized for promotion, while those who do not are being passed over. The policy sounds rational on its surface. It is not.

The problem is not that companies want workers to use AI. The problem is that adoption and competence are being treated as equivalent, when the research suggests they are structurally distinct. What these promotion mandates are actually measuring is behavioral compliance with a technology, not the development of any meaningful capability. The difference matters enormously, and organizational theory gives us the language to explain why.

Adoption Is Not Competence

Hatano and Inagaki (1986) drew a foundational distinction between routine expertise and adaptive expertise. Routine expertise describes the ability to execute known procedures in familiar contexts. Adaptive expertise describes the ability to respond effectively when contexts change. A worker who learns to paste prompts into ChatGPT and format the output for a weekly report has acquired routine expertise with respect to one tool in one workflow. That worker has not necessarily acquired the capacity to reason about AI-mediated tasks in general.

When companies tie promotion to AI adoption metrics, they are, in practice, rewarding routine expertise. The worker who logs the most AI interactions, produces the most AI-assisted deliverables, or completes the most internal AI training modules earns the career signal. But none of these metrics distinguish between a worker who has internalized the structural logic of AI-assisted work and one who has memorized a set of tool-specific procedures. This is the awareness-capability gap applied at the organizational level rather than the individual one.

The Measurement Problem Is Structural

Kellogg, Valentine, and Christin (2020) argued that algorithmic systems at work tend to create new forms of visibility and invisibility simultaneously. Certain behaviors become legible to management while others remain opaque. AI adoption mandates reproduce exactly this dynamic. The behaviors that are easy to track, such as tool logins, prompt counts, and completed certification modules, become the de facto definition of AI competence, not because they are valid proxies but because they are measurable ones.

This is not a new problem in organizational theory. Goodhart's Law, the principle that any measure which becomes a target ceases to be a good measure, applies with particular force here. Workers respond to the incentive structure, which means they optimize for the visible behaviors rather than the underlying capability. The result is that organizations can generate impressive adoption dashboards while the actual distribution of adaptive expertise remains essentially unchanged. The variance puzzle that motivates my own dissertation research, namely that workers with identical access show dramatically different outcomes, does not get resolved by mandating that everyone open the same tools.

What a Structurally Sound Promotion Criterion Would Look Like

The alternative is harder to administer but theoretically defensible. Gentner's (1983) structure-mapping theory suggests that genuine competence transfer occurs when learners have acquired relational schemas rather than surface-level procedures. A worker who understands why AI tools fail in certain contexts, what prompt structure affects output quality and by what mechanism, and how to evaluate AI-generated output critically rather than accepting it as authoritative, that worker has something transferable. Their competence generalizes across tool versions, platform updates, and novel task types.

Promotion criteria designed around this kind of competence would look different. They would involve performance on novel tasks, not just familiar ones. They would reward demonstrated judgment about when not to use AI tools, not just frequency of use. They would distinguish between a worker who can follow an AI workflow and one who can construct a new one when the familiar workflow breaks. None of these criteria are easy to reduce to a dashboard metric, which is precisely why organizations are not using them.

The Organizational Theory Implication

Rahman (2021) described how algorithmic management systems constrain worker agency through invisible structural rules rather than overt directives. AI promotion mandates invert this slightly: they make the structural rule explicit while leaving its content vacuous. Workers are told clearly that AI use matters. They are not told what kind of AI use, why it matters, or how to develop the capacity that would actually justify the career signal being attached to it.

The deeper issue is that organizations are attempting to coordinate around a competence they do not yet have the conceptual vocabulary to describe. Adoption is a visible proxy for an invisible thing. Until firms develop better frameworks for distinguishing adaptive from routine expertise in AI-assisted work, promotion mandates will reward the wrong behavior and leave the variance in actual outcomes unexplained.