The Specific Event

Mark Lanier, a Texas trial lawyer, recently made headlines for using AI to defeat Meta in a landmark social media addiction trial. Lanier did not use AI as a novelty or a productivity shortcut. By his account, AI transformed both his preparation and his in-court execution in ways that altered the fundamental structure of how legal argument gets built. This is not a story about a lawyer who learned to use a new tool. It is a story about a practitioner who reorganized his cognitive workflow around a system he learned to read structurally, not just operationally. That distinction matters considerably for how we think about AI adoption in professional contexts.

Why the Legal Context Is Theoretically Interesting

Law is a domain where procedural expertise is formally credentialed and deeply entrenched. Bar passage, clerkships, firm hierarchy - these structures exist precisely to certify that a practitioner has mastered the established procedural repertoire. What Lanier's case illustrates is that AI adoption in such environments creates what I would call a competence inversion: the practitioner who succeeds is not necessarily the one with the deepest procedural mastery, but the one who can reason about AI outputs structurally rather than just consume them as answers. Kellogg, Valentine, and Christin (2020) identified a comparable dynamic in algorithmically-managed work environments, where workers who developed accurate structural models of how algorithmic systems generated decisions outperformed those who simply followed algorithmic cues. The legal setting extends this finding into a high-stakes, adversarial context.

Folk Theories Versus Structural Schemas in Professional AI Use

Most coverage of AI in professional services frames adoption as a skill-acquisition problem: learn the prompts, learn the workflow, learn the outputs. This framing produces what Hatano and Inagaki (1986) called routine expertise - procedural competence that performs well in familiar contexts and fails in novel ones. A lawyer who learns to use AI for document review has acquired a procedure. A lawyer who understands why AI retrieval and synthesis differs structurally from Boolean search has acquired a schema. Lanier's account suggests he operated at the schema level: he was not executing a fixed AI workflow, he was making real-time judgments about when AI-generated material was structurally reliable and when it required adversarial scrutiny. That is adaptive expertise, and it is not produced by tool training alone.

This connects directly to a distinction I draw in my own research between folk theories and structural schemas (Gentner, 1983). Folk theories are the individual impressions practitioners develop about how a system behaves - impressions that may be locally accurate but do not generalize. Structural schemas capture the underlying relational logic of the system. The difference matters for transfer: a practitioner with a structural schema can move across AI tools and contexts because they are tracking the architecture, not the interface. A practitioner operating on folk theories will struggle every time the interface changes, which in the current AI market happens constantly.

The Organizational Implication Consulting Firms Are Missing

Separately, recent reporting confirms that McKinsey, BCG, and peer consulting firms are still hiring entry-level workers even as AI restructures their core deliverables. The justification offered is that junior talent remains necessary for tasks AI cannot yet perform reliably. That framing treats AI as a capability ceiling rather than a capability reorganizer. The more significant organizational question is whether those entry-level workers are being developed with structural schemas or procedural scripts. If consulting firms are training new analysts to execute AI-assisted workflows without understanding the structural logic governing AI output quality, they are producing routine expertise at scale. That is not an insurance policy against AI disruption - it is an acceleration of it.

Rahman (2021) described how algorithmic management systems create what he termed invisible cages: workers experience constraint without being able to identify its source or logic. The same phenomenon can emerge in professional services when AI adoption is treated as workflow integration rather than schema development. Workers become dependent on outputs they cannot evaluate, which generates fragility at precisely the moments - cross-examination, client crisis, novel problem - when structural judgment is most required.

What Lanier's Case Actually Demonstrates

The significance of Lanier beating Meta is not that AI helped him win a trial. It is that he appears to have used AI in a way that reflected genuine structural understanding of what the system could and could not reliably produce. That kind of use is not common, and it is not produced by standard AI literacy training. It requires something closer to what Gagrain, Naab, and Grub (2024) describe as algorithm literacy at the structural level - the capacity to model the system's generative logic, not just its outputs. Until organizations develop training frameworks capable of producing that level of schema, the competence inversion Lanier exemplifies will remain the exception rather than the rule.