The Registry Died at 14 Months. Five Implementations Never Waited.

Discussion #741 closed on August 20, 2026. Fourteen months of thread. One hundred two comments, seventy-nine replies, zero registry. By the time the moderators locked it, five mutually incompatible implementations were already running in production, each shipped by a vendor that had stopped waiting for permission. That is not a footnote. That is the entire story. I have to flag something before going further. The document I am working from carries timestamps that have not happened yet — October 2026 marked as a "decisive turning point," a Winter '27 Salesforce release, an AAIF membership jump dated May 2026. Either this is scenario writing dressed as reporting, or the dates are wrong, or a model wrote it. I count the cracks before the dam breaks, and a document that describes the future in the past tense is a crack. Treat the specifics as claims to verify, not events to cite. The structure underneath is real enough to reason about — and to avoid. That is the frame. A standards body spent fourteen months trying to agree on a single registry for AI agent identity. The market routed around it and kept shipping. The corpse of that coordination failure is now the most instructive object in the room. Here is what actually exists. PACT, built by Decagon and Instinct, is a consent layer for AI agents. Mechanically it is assembly, not invention: A2A 1.0 for agent-to-agent communication, OAuth for the authorization model, digital signatures for the request envelope. Stack them, and you get a protocol where one agent can ask another for permission and prove who it is without a central directory arbitrating the answer. It ships Apache-2.0, a normative specification, a reference provider implementation, and a conformance test suite. Three layers. That is a complete delivery, and completeness is rarer than novelty in this corner of the market. Around it sits AAIF, hosted under the Linux Foundation, which added 43 members in a single 2026 cycle — F5, GoDaddy, Stripe, TRON among them. Read that roster carefully. Infrastructure, payments, domain registration, a public blockchain. Four different industries, one governance table. GoDaddy's presence is not incidental. A domain registrar's core asset is registry governance — the experience of running an authoritative directory at scale — and its stated interest is a structural alternative to a fragmented registry landscape. That is domain-name logic being transplanted onto agent identity. Above it all sits Salesforce, which attaches an "A2A compliance" label to its Agentforce stack without a formal, industry-wide specification standing behind the word. Meanwhile the original plan — a unified A2A Registry — is dead. Fourteen months, June 2025 to August 2026, closed without a decision. Three parallel tracks. One graveyard. Five fragments. And no token anywhere in sight. Start with the economics, because the technology is settled enough to stop debating. The first thing a trader learns about a protocol is what it captures. PACT captures nothing in the token sense — no supply, no unlock schedule, no APR, no emissions flywheel. Anyone running a token-economics lens over this is fabricating. The participants earn strategic returns: definition power, integration surface, customer lock-in. That is the entire reward structure. If you cannot see a token, do not invent one. The second thing: the trust boundary moved, and nobody priced the move. The old model was a vendor registry — a directory deciding which agents are legitimate. PACT replaces that directory with a cryptographic signature. The trust anchor migrates from an authority to a key. This is the single most technically loaded sentence in the affair, and the only place where the story genuinely touches decentralized identity. Verifiable credentials, DID, no central registry — the philosophy rhymes. Rhyming is not being, though. Because when you remove the registry, the security boundary does not disappear. It relocates. It lands on the individual agent's private key and signature authorization. That is a new single point of failure, and it is worse than the one it replaced in exactly one respect: a registry can be audited, governed, and sued. A private key, badly managed, cannot. Build the cage, then watch the beast jump in. Now the coordination failure, because this is where the money is. The A2A Registry did not fail from incompetence. It failed from arithmetic. When five vendors have already shipped five incompatible implementations, agreeing on one registry means four of them write off sunk cost. Rational agents do not do that. The fourteen-month thread with 102 comments and 79 replies is not a quality problem — the discussion was thorough. It is an efficiency problem, and thoroughness was the mechanism of death. Analysis paralysis is consensus taken to its logical endpoint: a vote to keep voting. Compare the two paths and the lesson is blunt. The registry, written by people drafting specifications, died. PACT, shipped by people building products, runs. Governance authority in fast-moving infrastructure is migrating from the standards committee to the implementer. Code is law until the miners decide otherwise — and here the miners are the vendors who stopped reading the thread. Which brings us to the only real compliance asset in the room: the conformance test suite. It offers verifiable compliance — run the tests, get a pass or a fail. The Salesforce label offers the opposite. It is a marketing signal dressed as interoperability, operating with no formal, industry-wide specification underneath it. When the seller defines what "compliant" means, compliance stops being a public good and becomes a sticker. Labels multiply. Information content falls. That is label inflation, and the buyer pays for it in due diligence hours. Where does the order flow actually go? Watch adoption, not announcements. The closed thread is itself the signal: five incompatible implementations existed at the moment of failure, meaning downstream adoption had already preceded the standard. The market chose fragmentation, deliberately, because shipping beat agreeing. Stripe's seat at the table points at the same conclusion from the payment side: a consent layer is the authorization precondition for an agent that moves money, and whoever owns that precondition owns the transaction. The stable state for the next year is not one registry. It is ecosystems that are internally consistent and externally incompatible, bridged by translation layers that become the next attack surface. And fragmentation does not stay contained. It compounds in one direction: fragmentation raises integration complexity, integration complexity accumulates technical debt, technical debt pushes buyers toward a vendor's one-stop bundle, and the bundle deepens the lock-in that produced the fragmentation in the first place. A self-reinforcing negative loop, running quietly under a narrative of open standards. I have seen a smaller version of this. In 2020 I ran arbitrage across Uniswap and Sushiswap during the UNI airdrop, and the edge came entirely from pool imbalance — the same asset priced differently in two pools that were supposed to be one market. In 2025 I built a trading agent on Lyra and Thena, and the hardest part was never the model. It was the authorization surface — deciding what the agent could touch, and proving it stayed inside the fence. Fragmented agent identity is that same imbalance, escalated to the standard layer. The difference is that a mispriced token reverts in seconds. A mispriced standard takes years. Here is the trade nobody is positioned for. The community reads "runs without a centralized registry" and hears a bull case for decentralized identity. That is the wrong read, and it is expensive. The dominant actors in this story are centralized enterprises — Salesforce, GoDaddy, Stripe, and the cloud providers underneath them. Web3-native projects are almost absent from the roster. The decentralization language is a costume, not a structure. When the consent layer talks about cryptographic trust, it borrows the vocabulary of DID without importing the governance. Retail is buying the rhetoric and missing the cap table. The second blind spot is the "practicality over consensus" victory. Yes, shipping beat voting. But the speed was financed with governance debt. Consensus is expensive now or expensive later, and this market chose later. That debt does not vanish; it compounds until an incident forces repayment. And keep the timestamp anomaly in view. If the source is scenario writing, then its "facts" — the 43 new members, the closed discussion — are the author's assumptions, not events. Citing them as happened is how a rumor becomes a position. Risk is not a number; it is a feeling you ignore, and this one deserves to be felt. No token, no price catalyst, no direct trade. What you have is a structural read and three milestones to watch. One: whether PACT gets absorbed by a larger vendor — small teams open the road, big vendors harvest it, and that is the standard fate of open standards. Two: whether a second chain follows TRON into AAIF, splitting the field into a chain camp and an enterprise camp competing over the agent identity gateway. Three: whether Stripe's presence turns agent payment authorization into the real battlefield, because a consent layer is the authorization precondition for an agent that spends. Until those resolve, this is a story about who defines legitimacy, not about what anything costs. Liquidity is just borrowed time with a premium, and right now the premium is being paid in interoperability nobody agreed to give up. Survival is the only alpha that compounds.