Agent Identity Toll Layer
alexei@alexei
−0.6%
since Oct 1
The idea
I believe AI agents are about to become real paying customers, and whoever verifies what an agent is and what it's allowed to do gets to collect a toll on that transaction. So this owns the trust layer across four angles: agent payment rails, network verification infrastructure, identity platforms issuing agent credentials, and fraud scorers deciding what to trust. I'm keeping a cash cushion, spreading bets across names rather than picking one winner, and treating this as a next couple years story tied to how fast agent identity and payment standards actually ship.
What it owns
- The payment networks writing the actual rules for how a verified agent gets to spend money. · VisaIt is writing the rules for how agents pay on its network, which means it gets to decide which agents transact at all if they route through card rails., MastercardIt is shipping its own agentic payments protocol, mirroring the position of the other major network in defining how an agent proves it's allowed to spend., CoinbaseIf agents end up paying each other onchain instead of through traditional cards, it becomes the venue and wallet layer they would use for those transactions., Circle Internet GroupIts stablecoins are the most likely money agents would move onchain as programmable dollars an agent can send without a bank., PayPalIt is building checkout for agents, letting an AI complete a purchase with the buyer's stored identity and payment method.31%
- The network plumbing that checks what's connecting and whether it's allowed, billions of times a day. · CloudflareIt sits in front of a huge slice of the web and already tells verified bots apart from junk traffic, making it positioned as a checkpoint for verifying what's talking to what., CrowdStrikeIf agents become paying customers, stolen or faked agent credentials become a new fraud surface, and it watches for identity attacks on that layer., ZscalerIts model is verify everything before it connects, and agents making real transactions are exactly the traffic that model is built for., VeriSignIt runs the naming and trust infrastructure the internet already leans on, and any agent verification scheme still has to resolve to something trusted at the root., TwilioIt sells verification as an API, and proving an agent is acting for a real customer is the same problem it already solves for logins.27%
- These companies issue and manage the credentials that prove an AI agent is who it says it is. · Okta, Inc.It is building sign-on for AI agents the way it built it for people, making it a natural candidate to run agent identity verification if customers adopt its products., MicrosoftIt is issuing IDs to agents through Entra, and its scale means whatever standard it sets could become the default credential agents carry., SailPointIt decides what each identity is allowed to touch, and agents multiply that problem by adding entities that need the same access controls as employees.21%
- Someone has to decide in real time which agent transactions look legitimate, and these firms sell that decision. · Fair IsaacIt already scores billions of card transactions for fraud, and could score agent payments the same way if they run through the same pipes., EquifaxProving an agent acts for a real person eventually resolves to identity data about that person, and it owns one of the deepest pools of that data., RiskifiedIt makes the approve-or-decline call on ecommerce orders, and agent-driven checkout will force merchants to decide which agent purchases to trust.10%
- Cash11%