The White-paper Trap

The White-paper Trap

Agents are going to make payments. This future is already here, just not evenly distributed. Naturally then — given nearly half of all payments in the world are made with them — agents will need to be able to use card schemes, now enabled by platforms such as Visa Intelligent Commerce.

Now whilst this provided the much-needed mechanism for agents to actually use cards securely, in doing so brought to centre stage the underlying premise of all of payments that are now broken due to agents; the human is no longer necessarily there and present when the payment is being made. This presents a fundamental, 2 part problem:

  • Identity. Who is this? Who is it acting on behalf of?
  • Authorisation. What are you asking to do? Do you have authority to do it?

It is here that the Agents Payment Protocol (AP2) gestures at what they feel is the solution. Whilst it does a good job at clearly defining the problem space and the core requirements any solution would need to address, like most white-papers it largely then hand-waves at the solution; suggesting some decently elegant technical solutions to some of the problems, whilst overall failing to deliver a really concrete and complete vision for how it would really work in reality.

Such is the way of many white-papers.

When Sumvin joined Visa's Intelligent Commerce pilot at the start of the year, we knew we would need to address the same problem space, though with AP2 feeling lacking, we started design on an entirely new approach. Working from first principles — starting with identity — we went beyond the white-paper trap to design, deliver, and demonstrate our own agentic payments standard.

Sigil & Stamped Mandates is our answer to the problem AP2 defined, but failed to solve.

Sigil

Sigil is the Sumvin identity credential purpose built for the agentic-era. Regulatory-compliant to enable real finance use-cases and completely user sovereign, it provides proof-of-personhood for both a user, and their agents, where being able to distinguish between which one of the two is presenting it, is intrinsic to its cryptographic design.

"AP2 leaves credential issuance out of scope" is not a small omission. Issuance is the product in agentic commerce. Carrying authority cleanly and clearly, without a governed root, is like trying to rebuild the internet whilst leaving Certificate Authority governance as an "ecosystem problem". Great for "interoperability" theatre, somewhat lacking for getting a product to production.

Financial use cases bring financial regulation with it. We obviously chose Sumsub; industry leading identity verification right at the core of the agentic credential.

Stamped Mandates

When it comes to authorising stuff in a cryptographically provable manner, you don't need to look much further than what the crypto industry achieved with Intents. A phenomenally well put together spec — purpose built to ensure users understood what they were authorising — and given Sigil is a user sovereign credential issued on-chain, the obvious place for us to start. So we did.

A mandate is an authorisation to do a thing. Encoded into a structured message and then signed, it becomes stamped. At its purest, that's it.

Thanks to the beauty of fancy cryptography, whether this was signed specifically by the user or their agent is easy to prove, but critically, proven in a manner that demonstrates any agent belongs to the user. This simple mechanism unlocks the future we were promised; users can demonstrably authorise broad, overarching mandates, and agents executing them can distinguish themselves as they go about doing so. Multi-purchase baskets. Complex financial workflows. Disparate cross-domain work. All finally realised.

We wanted to be realise it to its fullest, so turning a Stamped Mandate into a first-class citizen of the worlds largest payments network made sense. Truly agentic Visa card payments, on your rails, or ours.

Avoiding the Trap

Sumvin published our technical thesis back in February. Since then we've worked with our design and then launch partners, to build the trust model the standards system is now converging on — cryptographically verifiable user intents and distinct agent authenticator — before EMVCo/FIDO wrote those requirement classes down.

Not by committee. With a small, mission oriented team. Fresh thinking, and a willingness to iterate at the same pace as the frontier

Understanding that the AI frontier has brought a paradigm shift in how technology is going to be built doesn't magically make the decades of financial infrastructure, technical debt, and entrenched thinking disappear overnight however.

We delivered the primitives, protocol, and platform the agentic frontier deserved. We're looking forward to working with our new set of partners as it becomes the one the world needs; orchestration of all of finance.