Should an agent be able to pay per call instead of holding a key?

Huginn agents all follow the same shape: watch a service, hold a credential for it, act on a schedule. The Plivo agents in flight for SMS and voice and the long-running agent suggestions thread show that shape keeps producing new agents, and every one of them adds another key to store, rotate and pay for up front.

BlockRun answers chat, image, video, music, speech and sound effects from one OpenAI-compatible base URL, and it also runs phone and voice, with each call settling in USDC over x402 rather than against an API key. An agent using it would work without a credential to seed. https://blockrun.ai

Would an agent written that way fit what the project accepts, or is a paid-per-call service something you would rather people wire up themselves? Happy to draft it in the shape of the existing agents and leave the decision with you.