Familiarity is the enemy: On why Enterprise systems have failed for 60 years
5 hours ago
- #knowledge-management
- #enterprise-software
- #ai-adoption
- Enterprise knowledge systems have consistently failed over 60 years due to buyers prioritizing familiar vendors, languages, and architectures over effective solutions.
- Familiarity is the primary selection criterion in enterprise software, driven by risk aversion and career protection, leading to massive financial losses and missed opportunities.
- The five failure modes include choosing familiar vendors (e.g., SharePoint), languages (e.g., Java over Clojure), buyer motions (avoiding outcome-based contracts), repeated failures (e.g., Cyc, expert systems), and AI stacks (e.g., RAG limitations).
- AI and large language models offer a third option: automatically generating structured knowledge graphs from unstructured content, breaking the historical trade-off between manual encoding and unstructured wikis.
- The case of Nubank demonstrates success through unfamiliar choices, using Clojure and Datomic to build a scalable, innovative banking platform, contrasting with typical enterprise decisions.
- A proposed anti-familiarity architecture includes Clojure, Datomic, graph-native design, entity resolution, deterministic AI harnesses, and sovereignty, aiming to address essential complexity in knowledge management.
- Four diagnostic tests (gap-analysis, entity resolution, time-travel, sovereignty) help evaluate if current systems are based on familiarity rather than correctness, highlighting common weaknesses in enterprise stacks.