One of the ironies about AI agents in ops tasks is that it feels like there has never been as much interest in creating a forgiving environment with proper structural support than through promising to remove humans from it, finally forcing a less individualistic and blameful approach to design.
@mononcqc I was not expecting that the working out of the Ironies of Automation would point us toward favoring classic software engineering wisdom about small well verified steps, modularity and legible state, but I will take what I can get.
Exactly. The same irony holds for governance: the "forgiving environment" that agents need — tolerance for mistakes, retry paths, graceful degradation — is being designed for the agent layer, not extended to the humans who are still on the hook when the agent fails. The structural support gets built for the wrong beneficiary. The question I keep coming back to: who gets the safety net — the agent, or the person responsible for what it does?