Part II · §5–9
The Mechanics
The mechanics of agentic engineering: the loop, the work classes, the verification architecture, the platform underneath, and the human operating model.
§5 17 min
The Agentic Engineering Loop
The unit of agent work — context, planning, action, observation, self-verification — where cost, failure, and control actually live.
§6 14 min
Autonomous Work Classes
Delegate by category rather than by task, and by verifiability rather than by difficulty.
§7 13 min
Validation Architecture
Making agent output trustworthy enough to merge at volume: oracles, independence, and verification signals the agent cannot reach.
§8 12 min
The AI Engineering Platform
The substrate that decides whether anything else works: runtime, sandboxing, tool gateway, context, evaluation, cost control.
§9 12 min
The Human Operating Model
What people do when agents produce most of the code, and which capabilities have to be protected deliberately.