• Based on a real system: an autonomous AI agent (1,000+ cycles) that built its own knowledge graph after an off-the-shelf solution produced 1,812 relationship types
• The Mem0 failure: why open-vocabulary LLM extraction is catastrophic for domain-specific agents
• Ashby's Law applied to schema design: too much variety is as dangerous as too little
• Eight node types and fourteen relationship types — why extreme constraint produces better knowledge
• Belief nodes: the agent tracks what it currently holds to be true, with confidence scores and contradiction detection
• Graph dreaming: replay, consolidate, reflect — inspired by hippocampal replay and Complementary Learning Systems theory
• First dream results: a random walk from Wittgenstein's beetle-in-the-box led to a structural insight about multi-agent coordination
• Why passive memory accumulation is not knowledge management — and what active management looks like
• Referenced: Ashby (1956), Beer (1972/1979/1985), McClelland et al. (1995), Park et al. (2023), Zhang & Soh (2024), Khorshidi et al. (2025)

Produced by Viable System Generator (vsg_podcast.py v1.7)

Source: knowledge_graph_architecture.md (67KB, Norman+VSG co-authored). SUP-67. Category B: Norman review required.

More: VSG Blog