An AI agent with a wallet full of stablecoins is useless if it doesn’t know what to buy or where to find it. That coordination problem—service discovery for autonomous software—is the bet behind Orthogonal, a startup that just closed a $4.3 million seed round led by Pantera Capital. The funding, announced this week, positions Orthogonal at the intersection of two accelerating trends: the proliferation of AI agents capable of executing financial transactions, and the growing patchwork of payment rails like x402 and Skyfire that give those agents the plumbing to spend. But plumbing alone doesn’t tell an agent which API to call when it needs compute, data, or inference. Orthogonal wants to be the yellow pages for machine money.
The Discovery Gap in Agentic Commerce
The current wave of AI agent payment infrastructure solves a narrow problem: how an agent transmits value. Coinbase’s x402 standard lets an agent pay for an API call in USDC on-chain. Skyfire and Payman connect agents to traditional payment networks. But none of these protocols answer a prior question: how does an agent discover which service to pay in the first place? Today, that discovery is hard-coded by developers. An agent built to fetch weather data calls a specific weather API because a human programmer wrote that endpoint into its configuration. If the agent needs a service its developer didn’t anticipate—say, a specialized legal document parser—it has no way to find one, evaluate its reliability, and negotiate payment. Orthogonal’s thesis is that this discovery layer must be built as dedicated infrastructure, not bolted onto payment rails after the fact.
How Orthogonal’s Architecture Fits the Stack
Orthogonal describes its product as “AI agent payment and service discovery infrastructure,” which suggests a two-sided marketplace architecture. On one side, service providers—API operators, data vendors, compute providers—register their endpoints, pricing, and service-level guarantees. On the other side, AI agents query the network to find services that match their needs, compare terms, and route payment. The payment itself likely settles through existing rails rather than a proprietary Orthogonal token, though the company hasn’t disclosed its settlement mechanism. The Pantera-led round, which included participation from other undisclosed investors, gives Orthogonal runway to build this registry and the agent-side SDK that makes querying it programmatic. The hard technical problem is trust: an agent needs to know that a discovered service will actually deliver what it promises, which implies some form of reputation system or cryptographic attestation.
Why VCs Are Betting on Agent Infrastructure Now
Pantera’s lead on this round fits a pattern. Over the past twelve months, venture capital has flowed into the agentic payments stack at every layer: MoonPay for identity, Skyfire for fiat on-ramps, x402 for on-chain settlement, and now Orthogonal for discovery. The logic is that no single company can own the entire stack, but each layer must exist for autonomous commerce to function at scale. Orthogonal’s $4.3 million seed is modest by crypto infrastructure standards, but it reflects a conviction that discovery is a distinct and defensible layer—not a feature that payment protocols can add as an afterthought. If that thesis holds, Orthogonal becomes a piece of critical plumbing that every agent SDK integrates by default, much as DNS became a default for human internet navigation.
Open Questions and Tradeoffs
Orthogonal’s success depends on solving a chicken-and-egg problem: service providers won’t list without agents querying, and agents won’t query without a critical mass of listings. The company will likely need to seed the registry with high-quality services and possibly subsidize early adoption. There’s also a standardization risk. If x402 or another payment protocol bakes service discovery into its own specification, Orthogonal’s standalone value proposition weakens. The counterargument is that discovery is a harder coordination problem than payment, and protocols optimized for settlement are poorly suited to solve it. Finally, the reputation problem looms large. A decentralized service registry without robust sybil resistance becomes a vector for spam and fraud, which would poison the well for agent adoption. How Orthogonal plans to handle reputation and curation will determine whether this is infrastructure or a proof-of-concept.