The missing chain of command

Give every agent a chain of command. Treating agent access as merely a working connection leaves unanswered who authorized a particular action, which policy applied, and what evidence remains. A useful chain of command has to answer those questions directly.

A chain of command has three properties, and all three have to hold at once. Someone has to be able to say, in advance, what an agent is allowed to do — not in the abstract, but for this session, these tools, this duration. Someone has to be able to say, at the moment of the call, whether it’s allowed — not “probably fine” but a real decision. And afterward, anyone asking “why did this happen” has to get an answer that traces back to the first two steps, not a guess reconstructed from scattered logs.

Humans authorize. Policy enforces. Evidence proves. Drop any one of those and the chain isn’t a chain — it’s a hope. An agent with broad standing access and no per-call decision has authorization without enforcement. A system that blocks calls but can’t explain why afterward has enforcement without evidence. Neither one answers “under what authority did this happen,” which is the only question that matters once something has already gone wrong.

The uncomfortable part is organizational: a team has to decide that “which agent did what, under whose authority” must be answerable on demand, not reconstructed only after an incident.