The agent commerce protocol war is real.

Six months ago there was no agent commerce protocol. Today there are six. Every payment network, every cloud, every AI lab is shipping one. ACP from Anthropic, UCP from a coalition, AP2 from Google, MCP from… also Anthropic. Mastercard's Verifiable Intent. OpenAI's Operator. Six standards, racing to be the standard.

It looks crowded. It is crowded. Most of these will not survive. The interesting question is which one wins, and why.

The asymmetry that matters

A successful protocol does one thing very, very well. HTTP moves bytes between machines. SMTP moves messages between mailboxes. Each was beaten by something simpler in its space and not by something more sprawling.

The agent commerce stack has six things that need to work: identity, messaging, oracle, reputation, settlement, and the agent runtime itself. Most of the protocols try to ship all six in one bundle. The one that ships the smallest useful unit first — and lets the rest of the stack compose around it — wins.

Why we're betting on small

Mumm publishes its primitives independently. Identity is its own RFC. Messaging is its own RFC. Oracle is its own RFC. You can adopt any one without committing to the others. We think this is how agent commerce actually gets built — incrementally, in production, by teams who don't have time to wait for the big-bundled standard to settle.

It also forces us to be honest. If our oracle is good but our messaging is bad, you can keep the oracle and use somebody else for messaging. The forcing function pushes every primitive to be useful in isolation.

The protocol war ends in interop

Three years from now most agents will speak two or three of these protocols and translate at the edges. The standards that win will be the ones whose interop story is honest, whose docs are clean, and whose reference implementations are open enough to fork.

We're trying to be that. We'll see.