The new problem: agents that spend on your behalf
AI assistants are moving from answering questions to taking actions — booking a plumber, hiring a translator, commissioning a developer. That is genuinely useful, but it creates an accountability gap. When a human books a service, they witness the work and judge whether it was done. When an agent books it, who verifies the provider, who confirms what was agreed, and who notices when delivery falls short?
Three things an agent needs to book safely
Verifiable provider identity
Before booking, the agent should be able to check, programmatically, that the provider is who they claim to be and holds the credentials the task requires — not just a star rating.
A signed record of the agreement
The agent needs a tamper-evident receipt of exactly what was agreed: scope, price, timing, and the guarantee cap. This is the evidence anchor if anything goes wrong.
A claimable, capped guarantee
A defined, machine-readable remedy means the agent can detect a failure against the agreed scope and either flag it for the human or initiate a claim within the rules.
Why machine-readability is the whole point
A guarantee buried in a PDF of legal prose is useless to an agent. For an AI to act safely, the guarantee has to be structured data: scope, cap, claim window, eligibility, and status — all queryable through an API. Then the agent can do real work: confirm before booking that a provider is verified, store the signed receipt, monitor the claim window, and surface a likely claim to its human the moment delivery diverges from scope.
The shift is from “trust the rating” to “verify the receipt.” Ratings are reputation; a signed, machine-readable receipt is proof.
How the Gera Action Warranty is built for agents
- Every eligible booking generates a signed action receipt exposed as machine-readable structured data via the Gera API.
- Provider verification is anchored in the Gera Services Passport, which an agent can check before committing to a booking.
- The 15 covered claim types, their caps, and their claim windows are all queryable, so an agent can reason about eligibility without parsing prose.
- Gera also publishes an MCP manifest and AI-plugin descriptor, so AI clients can discover and use the warranty surface directly.
The result: an AI agent can book a verified provider, hold cryptographic proof of the agreement, and act on a capped guarantee — the accountability layer that makes agent-directed real-world commerce safe rather than reckless.