The General Transaction Layer for AI Agents: Why Verifiable Agent Commerce Is a Special Case of Verifiable Agent Action, and What the General Case Requires

AI agents are beginning to act with real authority—to buy, approve, file, escalate, and decide on a person’s or an organization’s behalf. The payments industry has recognized that such actions require verifiable authorization and a tamper-proof record, and in 2026 moved to standardize exactly that for one verb: pay. This note argues that payment is a special case of a more general problem—any consequential agent action needs the same discipline—and that the general case has structure the payment case can leave implicit: pre-execution policy admission, outcome verification against declared criteria, and a portable, cross-implementation evidence record not bound to any single platform or rail. We describe the Trusted Cognitive Execution Architecture (TCEA) as a reference design for that general layer, map the agent-payment mandate chain onto it as a special case, and argue that the general agent-action governance layer is, as of this writing, named as a need by emerging standards work and not yet claimed by any published consortium scope. This note states the position; an open reference implementation and specification corpus follow.