Forty-two financial institutions. Zero production transactions. That ratio is the entire story of the authorization layer, and almost nobody covering it this week has bothered to write it down.
On October 6, a coalition unveiled Poppy — Personal Agent Protocol — with a design-partner list that reads like a roll call of the incumbents: Visa, Mastercard, Stripe, PayPal, Bank of America, Wells Fargo, JPMorgan, Plaid. Nine days later, Proof shipped x401, a protocol that repurposes the HTTP 401 status code from "who are you" into "prove that someone authorized you."
Both are drafts. Neither has an audit, a public repository, or a settled transaction behind it. And yet both have already been described as the infrastructure of agentic commerce.
That gap — between the weight of the names and the thinness of the evidence — is where the real reporting starts. Where the code meets the chaotic human heart, the press release always arrives first.
Let me explain the mechanics, because the foundational stuff matters more than the announcements — and I over-explain on purpose.
For two years the financial industry's answer to autonomous AI agents was a wall. Charles Schwab CEO Rick Wurster said it plainly: the firm blocked agent-driven account access because it could not tell a customer's assistant from an attacker's script. Defensible in 2024. Untenable the moment agent frameworks started transacting on their own.
Poppy's answer is a four-node delegation chain. Node one is a natural person verified to IAL2 — the NIST assurance level that requires a government identity document and a real human behind it. That person signs a mandate: scope, spending cap, expiry, counterparty. The mandate is cryptographically bound to node three, the agent, which presents its credentials to a merchant via OpenID4VP. Node four is the merchant, which performs independent verification and, in theory, closes the loop.
Read that chain twice. Every link leaves a verifiable artifact. In legal terms it is an authorization letter — the technical equivalent of agency law's power of attorney, the bedrock of who owes what when something goes wrong. That is genuinely well-engineered. It is also assembled almost entirely from parts that already existed: OAuth, OIDC, OpenID4VP, verifiable credentials. The innovation is assembly, not cryptography.
Now the part the announcements skipped.
Notice what both Poppy and x401 exclude from their core mandate: payment. Settlement is handed to x402, the Coinbase-born, Linux Foundation-hosted protocol that extends HTTP 402. This is a clean architectural separation and a catastrophic governance one. Authorization lives with a bank consortium; settlement lives with a neutral crypto foundation. When an agent is manipulated by a prompt injection into moving money inside its authorized limits — real signature, real agent, merchant verification passed, user intent hijacked — the dispute lands in the seam between two protocols governed by two different bodies, and each will be able to prove it did its own job correctly.
I have audited enough of these designs to recognize the pattern. In 2017 I ran Python simulations against forty-plus ICO whitepapers and learned that the failure mode is almost never the cryptography. It is the incentive to skip the check. Here the check is node four, independent verification at the merchant. The merchant pays for that verification in latency and engineering hours. The merchant does not eat the fraud loss. You do not need a game-theoretic model to see what happens next: verification gets simplified, then sampled, then quietly dropped, and the chain degrades into self-attestation with a compliance sticker on top.
Then there is the structural inversion nobody in this industry wants to name. IAL2 is a centralized chokepoint. Every compliant agent transaction resolves back to a KYC'd human with a passport scan. This is the Web2 identity stack domesticated onto Web3 payment rails — and it is being sold to crypto natives as progress.
Follow the money to where verification executes. Protocols write rules; they do not run checks. The checking happens at identity providers and at the edge — Plaid, Jumio, Okta and Persona on the credential side, Cloudflare and Akamai on the gateway side. When I spent a weekend in 2021 interviewing NFT artists about who really owns a JPEG, the lesson generalized: the entity that performs the act captures the value, not the entity that names it. Here the naming is free and the verification is billable. That is the quietest, most durable trade in the entire architecture.
One piece of engineering deserves scrutiny. HTTP 401 means unauthorized: supply credentials. x401 stretches that into supply proof of delegation, which quietly upgrades HTTP from an authentication protocol into an authorization protocol. Elegant on paper. Awkward against two decades of middleware, caching layers and WAF rules that assume 401 means what it has always meant.
Which brings us to the standard war, and the map matters. Three camps are forming. On the financial side: Poppy, Visa's TAP, Mastercard's parallel effort. On the AI platform side: Google's AP2 with Verifiable Intent, and Stripe's ACP. On the crypto side: x402. Look at Poppy's forty-two partners and note who is absent — no Google, no Anthropic, no Meta, no OpenAI. This is a pure financial coalition. And note that x401's marketing leans hard on broad compatibility with ChatGPT, Claude, Grok, Meta's Muse, MCP and AP2. Compatibility is what the weak advertise. The strong publish specifications.
Here is the sentence nobody wrote down: x402 decides how an agent pays; x401 and Poppy decide who may pay, how much, and on what terms. The high end of the value chain belongs to traditional finance. Crypto gets the pipes.
Historical base rates are unkind. Libra had twenty-eight founding members and never shipped. Coalitions move from announcement to production in eighteen to thirty-six months, when they move at all.
The consensus reading is that this is a defensive moat — banks protecting interchange fees from stablecoin rails. That is true and it is also the least interesting thing happening.
The contrarian read is that Poppy may be the most crypto-positive development of this cycle, precisely because it fails to solve what it claims to solve. Identity verification is a commodity; Plaid, Jumio, Okta and Persona already sell it. The unsolved problem — the one AP2's Verifiable Intent gestures at and x401 does not — is intent binding: proving not who authorized the payment, but that the agent executed the intent that was actually signed. Nobody has shipped that at scale. Whoever does owns the next standard, and it will not be a bank consortium, because banks are structurally bad at adversarial machine learning.
There is another blind spot. Forty-two direct competitors — Visa and Mastercard, Stripe and PayPal, Bank of America and Wells Fargo — sat down to write admission rules for the agent economy. Regulators call that self-regulation when the rules are open, and a group boycott when they are not. Poppy has published no charter, no voting structure, no admission criteria. The Department of Justice's non-intervention is not a verdict; it is a delay. Rewriting the ledger, one story at a time, means noticing which stories are absent — and governance is absent from this one entirely.
And watch the antitrust angle more closely than the token angle, because there is no token — no supply schedule, no unlocks, no value accrual. If one ever appears, it will be a governance wrapper around a boardroom, and the boardroom will not be on-chain.
So what do you watch? Not the partner count. Sideways tape is for positioning, not for chasing logos. Watch three signals: whether Poppy publishes an open charter, whether x401 lands in a neutral foundation, and whether any major standard formally adopts it as its identity layer. And watch the only metric that cannot be manufactured — a disclosed number of agent-initiated transactions.
Until then, treat forty-two logos as forty-two intentions. The ledger records the authorization. It has never once recorded whether the human meant it.