We don't send our entire knowledge graph to an LLM.
We build context specifically for the decision we're making.
Smaller context.
Better signal.
Lower cost.
More predictable results.
Context engineering has been a bigger optimization than changing models.
