The Specific Event

Gabrielle Boyle resigned from her position at the AFL three days before the organization activated Microsoft's AI assistant, Copilot, across its workforce. Her stated reason was direct: she had been told she could not opt out. This is not a story about one worker's personal preferences. It is a story about what happens when organizations treat AI adoption as a logistics problem rather than a coordination problem. The AFL's rollout assumed that access to a tool is equivalent to productive integration of that tool. That assumption has a long theoretical history of being wrong.

Access Is Not Competence

The Algorithmic Literacy Coordination framework I am developing for my dissertation makes a specific claim that is directly relevant here: platforms and algorithmically-mediated tools do not assume pre-existing competence, but most organizations that deploy them behave as if they do. The AFL's approach, mandatory adoption with no opt-out, treats Copilot as a standard enterprise software rollout. Install it, train people for an afternoon, move on. This is the organizational equivalent of what Hatano and Inagaki (1986) called routine expertise, procedural familiarity without adaptive capacity.

Boyle's resignation signals something more analytically interesting than a conflict about worker autonomy. It reveals what I would call an institutional awareness-capability gap. The AFL became aware that AI tools exist and that competitors may be using them. That awareness did not translate into a coherent theory of how competence with those tools actually develops. Kellogg, Valentine, and Christin (2020) documented this pattern in platform work broadly: organizations systematically misread algorithmic mediation as a neutral layer on top of existing work, when it actually restructures the work itself.

The Topology of Coercion

There is a structural issue embedded in the AFL's rollout that goes beyond this single resignation. Mandating adoption without allowing refusal collapses the distinction between having a tool and knowing how to navigate it. In platform contexts, Schor et al. (2020) identified how dependency relationships form precisely when workers lack exit options, and how that dependency actively suppresses the experimental behavior necessary for developing genuine competence. An employee who uses Copilot under duress is not building schema-level understanding of how the tool mediates their work. They are performing compliance. Those are categorically different outcomes.

This matters for organizational theory because it highlights a design failure that is becoming increasingly common. Firms are deploying AI assistants at scale while simultaneously eliminating the conditions under which workers could develop meaningful proficiency. Hancock, Naaman, and Levy (2020) argued that AI-mediated communication requires users to develop new mental models of how information is filtered, generated, and attributed. Mandatory rollouts with no deliberate schema-building component produce none of that. They produce workers who are technically "using AI" in the same way that someone who can start a car is technically a driver.

What the ISCA-ICAI Partnership Gets Right by Contrast

The same week as Boyle's resignation, the Institute of Singapore Chartered Accountants and the Institute of Chartered Accountants of India announced a formal partnership focused on responsible AI learning for accountants. The framing is different in a way that is not cosmetic. The word "learning" is doing real theoretical work there. Rather than treating AI adoption as a deployment event, the ISCA-ICAI initiative treats it as a competence-development process that unfolds over time. That framing is structurally consistent with what Gagrain, Naab, and Grub (2024) found in their analysis of algorithmic media use: algorithm literacy develops through structured engagement with structural features of the system, not through passive exposure.

The contrast between the AFL model and the ISCA-ICAI model is not about the pace of adoption. It is about what theory of learning is embedded in the organizational design. One treats competence as binary, present or absent based on access. The other treats it as developmental, requiring iterative engagement and structural understanding.

The Practical Implication for Organizations

Boyle's resignation will likely be read by most executives as a story about employee resistance to change. That reading misses the more important signal. When workers have enough schema-level understanding to accurately assess that a mandatory AI rollout will not produce meaningful benefit for their specific work, and then leave rather than comply, the organization has a measurement problem. It cannot distinguish between productive resistance, which carries real information about tool-task misalignment, and unproductive resistance, which reflects only discomfort with novelty. The AFL did not build the diagnostic capacity to make that distinction. Most organizations currently deploying AI at scale have not built it either. That gap is worth more analytical attention than the resignation itself.