Six to eighteen months after deploying AI tools, knowledge-intensive firms see a pattern that leadership can perceive but cannot name: throughput rises, deliverables proliferate — and the judgment of junior practitioners quietly stops developing.

This is the central management problem of the AI era, and the frameworks built for an earlier age cannot describe it. Nonaka and Takeuchi’s SECI model — the dominant theory of organizational knowledge for three decades — presupposes that every participant in knowledge work is a knowing subject with shared phenomenal context. Large language models break each of those presuppositions.