Insights
Essays and working notes, published on Substack. New posts appear here automatically.
Ask AI only for the precision you can really act on
Read on Substack
If nobody decides what "on-time delivery" means, your AI agent will.
Read on Substack
Everyone says "ontology." Agents need a knowledge layer.
Read on Substack
When an AI agent does the wrong thing, it either believed something false or wanted the wrong thing. The two call for different fixes.
Read on Substack
Organizations already pass intent down a chain, from board to desk. AI agents receive only the last link.
Read on Substack
Knowledge tells an AI agent what is true and what is allowed. It cannot tell the agent what you are actually trying to achieve.
Read on SubstackResearch and frameworks, free to download. Available in English and Japanese.
On the effectiveness and cost of grounding AI agents in an organization’s own knowledge
When an AI agent lacks an organization’s knowledge, it guesses, and its guesses look like answers. This paper sets out the evidence for building a governed knowledge layer — better accuracy on the same data, agents that know when to stop and ask, and lower cost — and how to start, one business decision at a time.
Subscribe on Substack. No spam, unsubscribe anytime.
