Observations on tacit knowledge at KMWorld 2026
As the first speaker in a series of three short keynote talks at KMWorld, Dave Clarke, CEO, Squirro, addressed what he called "The Amnesia Problem." Amnesia is about the loss of knowledge, particularly tacit knowledge. Clarke sees the crux of this problem as enterprise KM forgetting everything it learns The important question for KM is whether AI will become omniscient, will it know everything?
Traditional KM challenges include poor findability (now a solved problem, according to Clarke), taxonomy lag (AI is fixing this and it will be solved soon), tacit gap, and capture burden. The latter two are on their way to being solved. AI can compress some of the tacit knowledge capture burden by converting human experience into structured, compounding, queryable intelligence. Tacit becomes documented, created in a new space. Additionally, AI can detect discrepancies and anomalies in human interviews. Clarke recommended a KM checklist to get started on the journey from amnesia to enterprise omniscience.
Arvind Gopal, VP, product management & product strategy, eGain, sees knowledge as the operating manual for AI. He reflected on the Van Neumann architecture from the 1940s and compared to the computing paradigms of today. Language models in agentic systems combine knowledge (instruction) and data (structured and unstructured).
AI KnowledgeOps turns tacit knowledge from conversations and content into living knowledge for trustworthy AI. It’s a continuous cycle from capture demand to AI-ready content to synthesize content to organize and annotate to ground and test to orchestrate agents to deliver o all channels to learn from feedback and then it starts over again with conversations and content. Its AI Knowledge Hub uses natural, conversational language. It can do such actions as bulk changes, creating metadata, workflow processing, eliminating auding and versioning.
Gopal introduced its new eGain Trusted Knowledge Service, designed to make your knowledge AI-ready, on the platform you already have. It involves assessing how AI-ready your knowledge is, fixing content so AI can use it, proving whether your platform is the problem, and keep it trusted as it runs over time.
Knowledge graphs are not the solution; a disciplined Corporate Ontological Contract is, claimed Ken Stroker, principal solutions architect, Graphwise, in his talk on Grounding the Graph: Investing in Semantic Infrastructure. Stroker advised not having each AI team define its own meaning (what's a customer versus a client?) with nobody owning the whole.
Vectors have a place but they're not the solution. You can't extract an agreement from systems that disagree. Stroker strongly recommended establishing a corporate Taxonomy & Ontology Officer, a job might well fall under the Chief Data Officer. Treat projects like code. The ROI is in the discipline. Make your agreements deterministic, and future-proof them. Talk to your AI teams.
Companies and Suppliers Mentioned