The Specific Event

A story circulating this week describes what practitioners are calling "workplace paralysis" among software developers. The premise is straightforward: AI coding tools are being released at a pace that some developers find invigorating and others find genuinely disorienting. The story frames this as a cost of the AI coding boom - not in compute or licensing fees, but in human cognitive load and decision-making capacity. Some developers cannot determine which tool to adopt, which to deprecate, or how to build stable workflows when the underlying toolset changes faster than any workflow can mature.

This is not a story about AI being too powerful. It is a story about competence architecture failing under conditions of rapid environmental change. And it maps almost precisely onto a tension that organizational theory has not yet resolved.

Why Standard Change Management Misreads This

The conventional organizational response to technology-driven disruption is training. You identify a skill gap, you close it with instruction, and you restore productivity. This logic works when the target competence is stable. It fails when the target itself is moving. What the paralysis story describes is a situation where developers possess high domain competence - they can code, they understand software architecture, they can evaluate tools in isolation - but still cannot function effectively because the coordination problem is not about any single tool. It is about navigating a continuously restructuring landscape of options.

Kellogg, Valentine, and Christin (2020) make a related observation about algorithmic work environments: workers develop awareness of the systems they operate within, but awareness does not translate into improved performance outcomes. The developers described in this week's story are not unaware of AI coding tools. They are overwhelmed by the structural velocity of change itself. Awareness of the problem and capacity to respond to it remain decoupled.

The Routine Versus Adaptive Expertise Problem

Hatano and Inagaki (1986) draw a distinction between routine expertise and adaptive expertise that clarifies what is actually breaking down here. Routine expertise is procedural: a developer learns how to use GitHub Copilot, builds a workflow around it, and executes reliably within that workflow. Adaptive expertise is structural: a developer understands why certain tool architectures produce certain output behaviors, which allows them to evaluate new tools against principled criteria rather than starting from zero each time.

The paralysis problem is, at its core, a routine expertise trap. Developers who built their competence around specific tool procedures are finding those procedures deprecated or superseded faster than new procedures can be internalized. Developers who built their competence around structural understanding - what a model is doing when it autocompletes, where context windows create reasoning artifacts, how prompt framing shapes output distributions - are better positioned to absorb new tools without full workflow reconstruction. This is not a small distinction. It is the difference between organizations that compound capability across tool generations and organizations that restart their learning curve every six months.

The Organizational Accounting Gap

A second story from this week's news makes a related point from a different angle. The argument is that AI is already doing real work inside enterprises, but that work never reaches the ledger. The framing is that this is an accounting failure, not a security failure. I think that framing is partially right but underspecifies the mechanism. The reason AI work does not appear in organizational accounts is not just that accounting categories have not caught up. It is that organizations do not have a coherent theory of where AI competence lives.

If competence is treated as residing in individual workers, then AI tool adoption looks like a training expenditure that either succeeds or fails at the individual level. If competence is treated as residing in workflow architectures, then the relevant unit of analysis shifts to how tools are integrated into coordination structures. Rahman (2021) describes how algorithmic systems create what he calls an "invisible cage" - constraints that shape behavior without being visible as constraints. The developers experiencing paralysis this week are not just confused about which tool to use. They are operating inside a coordination structure that was built for a slower rate of environmental change than they currently face, and that structure is not visible in any budget line.

The Transfer Question

The practical implication worth examining is whether organizations can train for adaptive expertise in platform-mediated work environments, or whether that kind of expertise only develops through accumulated exposure. Gentner's (1983) structure-mapping theory suggests that schema induction - teaching people to recognize structural patterns across surface-level differences - is learnable and transferable. Applied here, this would mean training developers not on specific AI coding tools but on the structural features that distinguish tool categories: where in the reasoning pipeline a tool intervenes, what failure modes are characteristic of which architectures, and how output reliability changes as a function of task type.

Whether that kind of general schema training outperforms tool-specific procedural training in actual developer productivity is an open empirical question. But the paralysis story suggests the current default - procedural, tool-specific onboarding repeated every product cycle - is producing measurable dysfunction. Organizations that cannot close the gap between algorithmic awareness and algorithmic capability will keep paying for that gap in ways that, as one story this week correctly notes, never appear anywhere in their accounts.

References

Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive Science, 7(2), 155-170.

Hatano, G., & Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H. Azuma, & K. Hakuta (Eds.), Child development and education in Japan (pp. 262-272). Freeman.

Kellogg, K. C., Valentine, M. A., & 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.