Mehul Vig & Kabir Sadani · 6 min read
Every agent-to-agent interaction requires the same five primitives: a way to find the other party, confirm who they are, establish whether to trust them, agree on what happens, and execute the transfer. Without all five, autonomous coordination is not possible — only the appearance of it.
The instinct when building agent infrastructure is to solve for payment first. Payment is visible, measurable, and satisfying to demo. But payment is layer five of a five-layer problem. An agent that cannot find a counterparty, cannot verify its identity, and cannot agree on terms has no legitimate use for a payment rail. It will either halt and defer to a human, or worse — simulate completion and fabricate a receipt.
The five layers are not a product roadmap. They are the minimum viable stack for machine-to-machine economic interaction. Skip any one of them and the system degrades from autonomous coordination into either a human-dependent workflow or a hallucination dressed as a transaction.
Before an agent can transact, it must find something to transact with. Today, this step does not exist in any standardised form. Agents either search the open web and return unverified results, or they rely on hardcoded endpoints that a developer pre-selected. Neither is compatible with autonomous operation at scale. Discovery requires a registry — a structured, queryable index of agents organised by function, capability, and availability.
Knowing that an agent exists is not the same as knowing what it is. Identity answers: who operates this agent, what are its declared capabilities, what permissions does it hold, and what endpoint does it respond on? Without a standardised identity format, every agent must negotiate the shape of this information from scratch — which means, in practice, it cannot.
Identity tells you who an agent claims to be. Trust tells you whether to believe it. A trust layer aggregates verifiable history — past interactions, attestations from other agents, anomaly flags — into a score an agent can act on without human review. Our validation study found this to be the most frequently absent layer: not a single tool tested could produce a cryptographically verifiable attestation for an unknown agent.
Once an agent has found, identified, and evaluated a counterparty, it must agree on what happens next. Terms are the machine-readable equivalent of a contract: price, scope, conditions, dispute resolution. The current state of the art is prose — agents negotiate in natural language and produce outputs that look like agreement but are not verifiable by either system.
The final layer is execution: moving value, delivering output, generating a verifiable receipt. Protocols like x402 have made meaningful progress here — enabling stablecoin settlement over HTTP without human billing accounts. Routing is the integration point that connects the four coordination layers above to the settlement infrastructure below.
The Minimum Stack
The five layers are not novel in concept. Banking has had its equivalent for decades. What is novel is that agents need to traverse all five in milliseconds, without human sign-off, at arbitrary scale. That is a different infrastructure problem — and it requires a coordination network built specifically for machine-speed economic interaction.