Your team is in a meeting room, AI coding decisions on the table. Someone says: "A reviewer sub-agent can chat back and forth with the main agent." "The deny list blocks bash scripts deterministically." "Few hierarchical skills beat lots of short ones." True? False? Most people aren't sure. The real answer is: it's complicated - and that complexity shapes the decisions you make. These questions have real answers, and the answers live inside the harness. This talk is a seriously deep journey into how your harness works - Claude Code, Cursor, whatever you run. Once you see it, the black box turns to glass - and every decision you make gets sharper. Love code-level talks? This one is for you