We didn’t invent this. We put it to work.
Knowledge Strata didn’t invent the way organisational knowledge is classified. It draws on sixty years of thinking. Labour economics, philosophy, knowledge management. It turns that into a working tool for AI-era decisions.
We know more than we can tell.
A long lineage, made into one tool.
Four ideas sit underneath the framework. Drucker named the split between knowledge work and manual work. Polanyi showed that some knowledge can’t be fully written down. Nonaka and Takeuchi mapped how tacit and explicit knowledge convert. Collins split the tacit half into three distinct types.
Knowledge Strata adds the operational layer: four named layers, each with a plain test for telling them apart, and a clear role in deciding where AI fits.
Six thinkers, and where each one anchors the framework.
Peter Drucker
Named the split that starts it all: knowledge work is a different thing from manual work, and it needs to be managed differently.
AnchorsThe first distinctionGary Becker
Nobel, 1992Drew the line between general training, portable across any company, and organisation-specific training, worth something only where you are.
AnchorsWhy L1 / L2 differ from L3Michael Polanyi
“We know more than we can tell.” The founding observation that what a person knows is more than what they can put into words. The starting point for the whole tradition the framework rests on.
AnchorsThat L3b exists at allNonaka & Takeuchi
The SECI model: how knowledge moves between tacit and explicit forms in a four-stage cycle. Knowledge Strata doesn't replace it. The Extract loop sits inside it.
AnchorsThe Extract loopHarry Collins
Split the single idea of "tacit" into three working types: Relational (unsaid but tellable), Somatic (in the body), and Collective (held by the group). The typology is his. The action labels, Ask, Observe, Develop, and the diagnostic question are ours.
AnchorsAsk · Observe · DevelopBrynjolfsson, Li & Raymond
A study of 5,179 support agents. AI lifted productivity 14% on average. But 34% for novices and barely at all for experts. It compressed the documented learning curve (L3a) and left the experts' judgement (L3b) untouched.
AnchorsThe L3a / L3b split, measuredKnowledge Strata
Sixty years of thinking, put to work for the AI era. Four named layers, one plain test at each boundary, and a clear answer to where AI fits and where it doesn’t.
AnchorsThe synthesisThe edges are a feature, not a flaw.
Every classification produces cases that sit on the line. Good frameworks treat that as useful information. Cynefin (Snowden, 1999) goes as far as building a “confusion” zone into its model, making “I can’t yet tell which one this is” a valid first answer.
Knowledge Strata works the same way. When a piece of knowledge won’t classify cleanly, the boundary itself is telling you something: it sits where one kind of expertise turns into another.
In practice: two people arguing over whether something is L1 or L2 will, in settling it, name the exact judgement that separates general work from specialist work in their organisation. That naming is the point.
See the lineage put to work.
A full annotated bibliography is in preparation. In the meantime, the framework itself is where the thinking lands. Start with the four layers.