Distill
The framework applied.
Current signals read through Knowledge Strata. Each one shows the framework at work: what it reveals about a live market signal, and which layer the signal illuminates.
This Week’s Distillation
Tacit knowledge is your next competitive moat.
"The real differentiator is not the data or even the models, but the tacit knowledge embedded in the judgement of their people." Tung and Roussiere argue that organisations failing to operationalise tacit knowledge face escalating costs: brittle AI systems, inconsistent decisions, vanishing expertise and stranded investments.
What the framework reveals
The framework reveals that the five actions Tung and Roussiere propose — map tacit knowledge, codify without flattening, build a semantic layer, enable human-AI collaboration, align executives — are the L3b extraction pipeline without the operational classification that makes the methodology actionable. Their automaker case study (master engineers whose reasoning was captured into AI agents reviewing design archives) is a textbook Observe extraction. The framework adds what the article leaves implicit: not all tacit knowledge responds to the same extraction approach, and the diagnostic question ("can they explain it?") determines which one to use.
A peer-reviewed management journal from UC Berkeley has independently arrived at the same conclusion as the Knowledge Strata diagnostic — tacit knowledge is the defensible layer, and organisations that fail to extract it before AI adoption matures will find their systems brittle and their investments stranded.
Recent Distillations
Amazon fired 16,000 engineers. The AI rebuild broke everything.
The framework exposes that documented artefacts (L3a) lack the tacit understanding (L3b) that produced them. Amazon had the code. It did not have the judgement of the people who wrote it. The layer pattern shows that automating L1/L2 work without first extracting L3b creates a system that runs but cannot be safely changed.
Nobody knows what you're worth anymore.
The framework reveals that the chain where production signified effort signified expertise is broken because L1 generation is now free. This creates a knowledge-layer lemons problem — the market cannot price what it cannot see. A second pattern follows: AI removes the L1 tasks that were the training pathway to L3b, potentially widening the gap between what organisations need and what the next generation of workers can demonstrate.
"Hostages, not customers" — three patterns of AI defensibility.
The framework reveals three distinct defensibility patterns in Horowitz's three dynamics: Hierarchical (the regulatory gate determines entry, AI determines profitability inside it), Independent (channel difficulty is the moat, unrelated to knowledge type), and Collapsed (L1 products face existential pressure as model capability rises). Defensibility and knowledge layer are independent dimensions — a business can hold L1 knowledge and still be defensible, or hold L3b knowledge and still be vulnerable, depending on which pattern applies.
97% of enterprises run AI agents. 12% have governance. 40% will cancel.
The framework reveals the 40% cancellation rate as the market's forced correction for organisations that deployed AI agents at L1 (capability) without the L3a governance (documented rules, boundaries, escalation protocols) or L3b judgement (who decides what the agent should do when the rules don't cover it) to operate them safely. The governance shortfall is a Context Debt problem: the knowledge needed to govern the agents was never captured because nobody asked for it during the capability sprint. The six critical gaps the article identifies — absent centralised control, late-stage governance, missing decision traceability, no policy-as-code, unclear human-in-the-loop boundaries, role confusion — each map to a specific layer failure in the framework.
"AI thinks. Workflow acts." ServiceNow's CEO names the Three Systems.
The framework reveals that McDermott's positioning makes the vendor landscape legible: System of Record (where data lives), System of Intelligence (where AI reasons), System of Action (where decisions execute). ServiceNow captures L1 through L3a cleanly but has no mechanism for L3b. Every major vendor occupies a slice of the knowledge stack, none captures all four layers, and the gap is always at L3b — tacit organisational knowledge that no platform can reach.
Archive
Week of 6 January 2025
The Karpathy Moment Is Spreading Beyond Coding
This isn't just about coding. The pattern Karpathy describes — feeling behind despite being an expert — will ripple through every knowledge profession. The question isn't whether AI will commoditise your baseline work. It's whether you're building advantage at Layer 3 before it does.