I lead Safety and Telematics engineering at Samsara and build with coding agents hands-on. These essays draw on my independent work on All Is Vanity: how I set direction, use tools to iterate, and require evidence before accepting a change.

The work is about getting more useful capacity from a system. That includes the people, the agents, the architecture, and the feedback loop around them.

  1. AI gives me speed. My job is to give it direction. — Choosing an observable outcome and keeping parallel work aimed at it.
  2. The guardrails that let me run multiple AI agents — What an early eight-agent failure taught me about ownership and recovery.
  3. What an AI agent has to prove before I call it done — From a failing reproduction to the intended runtime state and a reviewable result.
  4. I build tools so the next iteration costs less — Editors, scenarios, shared contracts, and the economics of a shorter feedback loop.
  5. When I stop an AI agent and start a new session — Distinguishing useful exploration from repeated work, and preserving the right context.
  6. 2,161 commits later, here is what I measure — Activity, accepted results, and a proposed scorecard for AI engineering investments.

Repository snapshot and counting method · About Anthony · Projects

These are personal engineering observations. Repository figures describe my independent project, not employer productivity results.