From Old Wine to AI‑Native Teams: The Truth of Ontology Governance in AI

1. Ontology: A Repackaged Old Wine

Guest Zhang Sensen opened the discussion by likening ontology to “new wine in an old bottle,” arguing that it is essentially a layered data‑modeling approach that extends traditional dimension tables, code tables, and master‑data management. He emphasized that the real difficulty lies not in the technical construction of classes, attributes, or relationships, but in governance: deciding who has the authority to define and maintain ontology elements, and who arbitrates conflicts or erroneous model outputs.

2. The Modeling Paradox: Business Pain vs. IT Effort

Host Zheng Yan observed that, contrary to intuition, modeling is more painful for the business side than for IT. Early attempts relied on ad‑hoc solutions such as LangChain‑style skill pipelines, but the rise of large models changed how models are consumed, widening the gap with earlier ontology products like Palantir’s. Despite the shift, fundamental data‑engineering challenges—massive graph retrieval, real‑time materialized views, and multimodal data ingestion—remain hard problems that AI alone cannot solve.

3. Two “Diseases” of Enterprise AI Transformation

Zhang identified two common corporate mindsets: (1) treating AI as a cure‑all for existing business bottlenecks, and (2) adopting AI out of competitive pressure without a clear problem to solve. He warned that this “wish‑fulfillment AI” often leads to workforce reductions without genuine process redesign, and that companies risk losing competitiveness if they merely overlay AI on unchanged workflows.

4. Highlight Projects Conceal Massive Technical Debt

Ma Jinlong highlighted that rapid AI pilots generate “highlight projects” that accumulate technical debt far exceeding that of traditional software or data‑governance initiatives. He illustrated this with a case from the automotive sector where leadership demanded quick AI results despite weak data foundations, leading teams to chase low‑hanging‑fruit AI coding solutions that mask deeper governance shortcomings.

5. Steam Engine vs. Electric Motor: Organizational Inertia

Using a historical analogy, Zheng argued that swapping a steam engine for an electric motor does not transform a factory unless the underlying organization changes. He noted that many enterprises are willing to change but lack a clear roadmap for rebuilding processes from scratch, especially when legacy business must continue running.

6. Building an AI‑Native Organization from the “Baby” Stage

Ma described his team’s experiment in creating an AI‑Native organization. They adopted three principles: (a) all documentation must be AI‑readable, preferably in Markdown; (b) certain domains such as finance, legal, and IP remain human‑centric, with AI only augmenting efficiency; (c) the organization remains flexible, acknowledging that models and agents will evolve, requiring periodic reassessment of workflows.

He gave a concrete example where a single “AI‑augmented” engineer handled end‑to‑end project delivery, covering business liaison, requirement analysis, and implementation, while humans focused on validation and decision‑making. However, he cautioned that the real challenge shifts from building solutions to identifying the right problems to solve.

7. No Silver Bullet, but the Journey Must Begin

The panel concluded that there is no universal answer for AI adoption, yet enterprises must start moving forward. Consensus emerged that ontology governance is fundamentally a management issue, organizational inertia cannot be overcome by merely changing technology, and small‑scale, “baby‑state” experiments are the only viable path for many companies to become truly AI‑Native.