Three unrelated announcements from the past week show the same underlying shift: the identity problem for AI agents is being broken into distinct, composable layers instead of solved as one blob. 1Password shipped a zero-exposure integration for Claude. When Claude needs to log into a site on a user's behalf, 1Password itself injects the credential into the page after biometric approval — Claude never sees the password or one-time code. As 1Password's CTO put it, the goal is to let a user give an agent permission to use a credential without letting the agent see it. That's a narrow claim: authentication happens, but the agent's context never touches the secret. Separately, Dai Nippon Printing rolled out a digital identity function for AI agents making proxy purchases, built on its CATRINA distributed ID platform and verifiable credentials. The stated goal is an environment where only AI agents authorized by a specific user can transact, with the system expressing whose proxy the agent is acting as and under what authority — addressing the authorization question rather than the authentication one. And the Linux Foundation completed the operational launch of the x402 Foundation, the open-governance body now stewarding Coinbase's x402 protocol for internet-native payments over HTTP. The protocol lets AI agents, APIs, and applications send and receive payments as part of normal web traffic, with 40 member organizations — including AWS, Google, Visa, Mastercard, Stripe, and Coinbase — already signed on. That's the transaction layer: once an agent is authenticated and authorized, x402 is how it actually moves money. None of these three projects talk to each other directly. But laid side by side, they map cleanly onto separate concerns: who is the agent authenticating as (1Password), what is it authorized to do on whose behalf (DNP/CATRINA), and how does the actual transaction settle (x402). That separation matters more than any single feature. A monolithic "agent identity" product would bundle authentication, authorization, and settlement into one vendor's black box. Splitting them into interoperable layers — credential brokering, verifiable-credential-based delegation, and an open payment protocol under neutral governance — is what lets a downstream service verify a chain of trust without prior coordination with every vendor an agent might touch. It's slower and messier than a single unified stack, but it's also the only version of this that scales past one company's roadmap.