The Shift Nobody Is Talking About

A quiet but significant transition is underway inside large law firms. After two years of licensing AI tools, firms are now building proprietary systems around something more defensible than software: what only they know. Recent reporting confirms that firms are constructing AI infrastructure specifically designed to capture, index, and operationalize institutional experience accumulated over decades of client work. This is not an AI adoption story. This is an organizational theory story about how tacit knowledge becomes encoded capability, and what happens when that encoding succeeds.

The Tacit-to-Explicit Problem

Organizational theorists have long distinguished between knowledge that can be written down and knowledge that lives in practice. The classic problem is that expert judgment in fields like law does not transfer cleanly to procedure manuals. A senior partner reviewing a contract does not follow a checklist. She pattern-matches against thousands of prior engagements, regulatory shifts, client-specific risk tolerances, and court-specific tendencies. That process is largely invisible, even to the expert performing it. When law firms say they are building AI around their most valuable asset - experience - they are attempting something that has historically resisted formalization.

This connects directly to a distinction I work with in my dissertation research between folk theories and structural schemas (Kellogg, Valentine, and Christin, 2020). Folk theories are impressionistic, individual, and non-transferable. Structural schemas capture the underlying logic of a domain in ways that can be taught, applied, and generalized. The firms attempting this transition are not just digitizing files. They are attempting schema extraction at institutional scale, and that is a fundamentally different technical and organizational problem.

Why Routine Expertise Fails Here

Hatano and Inagaki (1986) draw a distinction between routine expertise, which is optimized for stable, predictable environments, and adaptive expertise, which allows practitioners to respond effectively to novel conditions. Legal work, particularly in high-stakes transactional and litigation contexts, is an adaptive expertise domain. The conditions change. Regulations shift. Counterparties behave unpredictably. Courts diverge in interpretation.

This creates a specific risk in what law firms are building. If the AI systems they construct encode routine patterns from historical cases without capturing the structural reasoning that generated good outcomes, firms will have produced expensive brittle systems. The systems will perform well on cases that resemble their training data and fail on precisely the cases where experienced judgment matters most. This is not a speculative concern. It is a known failure mode in organizational knowledge management, and the law firm context amplifies it because the cost of failure is not a bad YouTube video but a lost case or a missed liability.

The Competitive Moat Is Real, But So Is the Asymmetry

There is a legitimate competitive logic to what these firms are doing. If institutional experience can be encoded effectively, then the firm's historical case portfolio becomes a durable asset rather than a depreciating one. Smaller firms without decades of documented precedent are structurally disadvantaged in this race, regardless of the AI tools they can purchase. This maps directly to the variance puzzle that motivates my own framework: platform workers with identical access to tools show dramatically different outcomes (Schor et al., 2020). Access to AI tooling is not the differentiator. The differentiator is the quality and structure of what gets encoded.

Rahman (2021) describes how algorithmic systems function as invisible cages, shaping behavior through constraints that workers can feel but not always see. Law firms building proprietary AI systems are, in effect, constructing those cages internally and intentionally. The question is whether the cage encodes the best reasoning of the best practitioners, or whether it encodes the average behavior of whoever generated the most documentation.

What Organizational Theory Predicts

The theoretical literature on organizational knowledge suggests the outcome will depend on how firms handle the encoding process itself. Organizations that treat knowledge capture as a technical problem, essentially a data pipeline, will produce systems that reflect volume over quality. Organizations that treat it as a social and epistemological problem, asking which knowledge matters, who holds it, and how it can be made legible without being flattened, have a better chance of producing adaptive rather than routine AI capability.

The firms reporting the deepest progress on this transition are, notably, involving senior practitioners directly in system design rather than delegating it to technology teams. That organizational choice is not incidental. It is the mechanism. The schema induction problem cannot be outsourced to engineers who do not hold the expertise being encoded. The firms that understand this are not just building better AI. They are running a serious experiment in organizational cognition that the rest of the knowledge-work economy will eventually have to replicate or fall behind.

References

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

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

Schor, J. B., Attwood-Charles, W., Cansoy, M., Ladegaard, I., and Wengronowitz, R. (2020). Dependence and precarity in the platform economy. Theory and Society, 49(5-6), 833-861.