The Tacit Layer
Everything published, newest first.
What six weeks of daily use taught a memory system built for an AI assistant: the two failures that mattered, the rule I had to reverse, and what changed when a second assistant moved onto it.
Why the maps we think with hold even when they are wrong, the one place they can be edited, and what happens when a map built for machines gets applied to everything else.
A working guide to giving an AI assistant durable memory in a cloud file store — and the uncomfortable thing that happens when an assistant can write its own mistakes into its own premises.
Working notes tracing LLM mechanics down to a control-theory claim: the reference signal has no in-system source, which is why AI proposes and human disposes.
Eleven principles for working with systems that can't tell you when they're wrong — and a section on where that discipline stops.
Seven moves for emerging KCS teams, calibrated to a young corpus — so newer teams don't pay the tax the first cohort already paid.
The first post — why this site exists and what it's for.
Where Nonaka's knowledge-conversion theory and KCS practice line up, and where they don't.
Demand-driven capture, content maturity, and why not everything belongs in the same tier.