This is a viewpoint on how we represent know-how in AI systems: not just what an organization knows (knowledge graphs, RAG, etc.), but how it does repeatable cognitive work.
Web version: <https://knowhowgraph.com/> — TL;DR below. Feedback from people building agents / AI workflow tooling especially welcome.
# Viewpoint: The Know-How Graph
Declarative, Repeatable AI Workflows as Shared Infrastructure
TL;DR
Agents are great at solving new problems, terrible at doing the same thing twice.
We argue that repeatable AI workflows should complement agents: written in a declarative language that both humans and agents can understand, reuse, and compose. These workflows become tools that agents can build, invoke, and share to turn repeatable cognitive work into reliable infrastructure.
At scale, this forms a Know-How Graph: a network of reusable methods that become shared infrastructure.
Agent memory: knowledge graphs aren't enough, we need a know-how graph · HackerTrans
Web version: <https://knowhowgraph.com/> — TL;DR below. Feedback from people building agents / AI workflow tooling especially welcome.
# Viewpoint: The Know-How Graph
Declarative, Repeatable AI Workflows as Shared Infrastructure
TL;DR
Agents are great at solving new problems, terrible at doing the same thing twice.
We argue that repeatable AI workflows should complement agents: written in a declarative language that both humans and agents can understand, reuse, and compose. These workflows become tools that agents can build, invoke, and share to turn repeatable cognitive work into reliable infrastructure.
At scale, this forms a Know-How Graph: a network of reusable methods that become shared infrastructure.