Domain-Native Development: A Mekiki Framework for AI-Assisted Knowledge Work

When experienced professionals retire, organisations lose not their documented procedures but the accumulated judgement that determines how those procedures are applied. This paper proposes a two-component framework—termed the *mekiki* framework, after the Japanese concept of expert discernment (目利き)—for understanding how artificial intelligence (AI) makes this otherwise invisible knowledge structure observable. The framework identifies two conditions governing whether tacit knowledge can be successfully externalised into formal artefacts—software, documents, analytical tools: *externalisation cost* (the technical barriers to articulating and implementing knowledge) and *specification cost* (the barrier constituted by the domain expertise required to determine what should be built, how quality should be judged, and what should be excluded). Drawing on a revelatory case study in which a domain expert with no programming experience developed and publicly deployed a web application in ten working days using AI tools, the study shows that externalisation cost drops dramatically under AI assistance while specification cost persists unchanged. A domain-ablation experiment, in which two independent AI systems produced technically competent but domain-inappropriate applications from the same dataset without expert guidance, provides supporting evidence for the separability of the two components. Grounded in the SECI knowledge management tradition, the framework offers organisations a structural vocabulary for identifying which knowledge components are at risk of irreversible loss through generational transition—and which can be preserved through AI-mediated dialogue while retiring experts remain available.