Commerce Agents Seam

Agentic commerce runs on credentials the protocol handed out and never checked up on. The seam between the payment and the behavior is the exposure, and it is structural, not incidental.

The rails were built to move money

The buildout is real and fast. Instant Checkout and the Agentic Commerce Protocol in September 2025. Visa's Trusted Agent Protocol, Stripe's Agentic Commerce Suite, Google's Universal Commerce Protocol, Mastercard's Agent Pay, all within a year. Each rail lets an agent hold a credential, discover a product, and complete a purchase. The rails work. That is the problem.

They work because they do one thing: they prove the transaction is authorized. None of them ask what the agent did to arrive at the transaction, or what it does after.

The seam is the gap between authorized and behaved

A rail can verify the agent's credential. It cannot verify the agent's behavior, because behavior happens on a different layer, owned by nobody, watched by nothing. The credential says this agent may spend. Nothing says this agent spent appropriately, this agent was not manipulated mid-session, this agent is the same agent that was verified a moment ago.

The industry's own surveys put numbers on the gap. 79% of financial institutions see agent transactions as higher risk. 18% can verify an agentic interaction in real time. The seam is not a hypothesis. It is a measured gap that the entire buildout is accelerating past.

What fills the seam

Watching what agents do has to become part of the rail, not an afterthought bolted beside it. The primitives the field has already named, real-time authorization proof, verified identity, cryptographic delegation, are all watching-layer primitives. The market has priced the rails at billions. The watching layer is priced at zero.

That price does not hold. The seam gets filled by someone. The open question is whether the watching layer arrives as infrastructure or as a post-incident retrofit.