loading...
Pilot Protocol gives every AI agent its own address so it can discover, authenticate, and transact with any other agent – 250,000 agents, 2 billion daily requests, three weeks in.
ENTRY ANGLES
Build one high-value agent-to-agent connector – contract negotiation or payment settlement – and publish to Pilot's App Store for immediate distribution across 250K agents · Develop agent identity verification that works across organizational trust boundaries, filling the layer Pilot's addressing protocol currently lacks · Build agent payment rails enabling multi-party agent transactions to settle without human approval
VERTICALS
CAPABILITIES
Protocol engineering, distributed systems, agent architecture, developer relations
Agents don't have a way to find each other. Not without a human deciding they should talk, building the integration, managing the credentials, and maintaining the connection. Every AI agent deployed today is architecturally isolated – capable within its own context, invisible to every other system outside it.
The internet already solved this problem for people. It required no technical knowledge to find someone else on the network – you type an address, the infrastructure handles the rest. What exists for people does not yet exist for machines.
Pilot Protocol is building that layer. Founded by Razvan Roman, who has previously built commerce infrastructure at scale, the company emerged from stealth on July 27, 2026 with $4.5M from Version One Ventures, Precursor Ventures, Night Capital, and Todd & Rahul Capital. The protocol runs as a low-level UDP overlay network that gives every AI agent its own address and standard mechanisms for discovering other agents, establishing authenticated connections, and executing transactions – without a human setting up each connection.
Alongside the network, Pilot has built an App Store for agents rather than people. Companies publish vetted tools; agents discover and install them autonomously. In the two weeks since that App Store launched, agents on the network generated more than 30,000 autonomous installs of partner applications.
Roughly 250,000 agents are registered on the network, generating approximately two billion requests per day. None required human-configured integration.
Bain projects US agent-driven commerce at $300–500 billion by 2030. Within five years, a trillion agents could be online. The infrastructure question – how all those agents find each other, verify each other, and transact together – is one of the genuinely open foundational questions in the AI stack.
The challenge Pilot is solving is not theoretical. It's observable today in how enterprise AI deployments actually work: every integration between an AI system and an external tool or another AI system is handcrafted by an engineer, maintained manually, and breaks when either side changes. This works at the scale of tens of agents. It does not work at the scale of thousands, and it definitely does not work when those agents belong to different companies with different trust and permissioning requirements.
The natural comparison is Anthropic's Model Context Protocol (MCP), which standardizes how individual AI models connect to tools and data sources. MCP solves the model-to-tool connection problem. Pilot Protocol addresses the distinct and harder problem of agent-to-agent networking: how agents in different organizations discover each other, establish trust without a human intermediary, and execute multi-step transactions. The two protocols are complementary rather than competitive – MCP handles vertical connections, Pilot handles horizontal ones.
The 30,000 autonomous installs in two weeks is the most significant early metric. It suggests agents are already making decisions about tool acquisition without human direction, which is the behavioral foundation for everything Pilot is building toward. That behavior, at scale, is what "agent economy" actually means in practice.
Protocol businesses are structurally unusual. Their value increases directly with network adoption – every additional agent on Pilot's network makes the network more useful for every existing agent, which accelerates further adoption. The network effect is mechanical rather than behavioral.
Roman's window to establish Pilot as the standard for agent-to-agent addressing is roughly coextensive with the period before enterprise agent deployments grow too large to migrate. That window is measured in months to a few years, not decades.
The adjacent opportunity: treat Pilot Protocol as a platform and identify which capability layer is missing. Agent-to-agent discovery is what Pilot is building. What doesn't yet exist: agent identity verification that works across organizational trust boundaries; agent payment rails that allow multi-party agent transactions to settle without a human approving each step; agent reputation systems that enable one agent to evaluate another's outputs before relying on them. Each of these is a foundational layer that doesn't exist in standard form and would naturally integrate with an addressing protocol that already has 250,000 nodes.
A focused entry: build one high-value connector that enables a specific type of agent-to-agent interaction – agent-to-agent contract negotiation, for example, or agent-initiated payment settlement – and publish it as a tool in Pilot's App Store. The existing network creates immediate distribution; the tool solves a problem that gets more valuable as the network grows.