FILTER · TAG

#AI Agents

2entries · clear ✕

Journal

Does Graphify Actually Cut Token Usage?

RB-000707.12.2026

I committed a 15 MB knowledge graph and forced every agent to consult it before reading code, on the promise it would save tokens. So I ran a controlled test on a real task. It saved about 5%—but only by returning a worse, incomplete answer. Here is what I measured and why I pulled the whole thing out.

Read →
Journal

Don't Build the Tool. Deliver the Outcome.

RB-000506.17.2026

A response to Sequoia's "Services as the New Software." The next durable businesses won't sell another piece of software that races the next model. They'll deliver the actual service, measured in real outcomes like recovered revenue, audit findings, and savings found.

Read →