About Mandate Labs

What triggered this

I spent years building card issuance and processing infrastructure — a Mastercard principal membership, BaaS platform, the full stack from BIN sponsorship to settlement. That work gave me an intimate understanding of how authorization decisions actually flow: the millisecond-level handshake between agent, issuer, network, and merchant that determines whether money moves.

Then I watched AI agents start requesting those authorizations. Not hypothetically — in production. And I realized that the entire trust architecture of the financial system was built around a single assumption that was about to break: that a human made a conscious decision before every transaction.

There was no verification layer for the decision itself. Identity? Yes. Creditworthiness? Yes. But whether the reasoning behind the transaction was sound, aligned with the principal's intent, operating within stated constraints? Nothing. That gap is what started Mandate Labs.

What worries us

AI agents are going to handle trillions in commerce. That's not a question anymore — it's a trajectory. The question is what happens in the absence of verification infrastructure.

Without it, institutions face an impossible choice: block agent-initiated transactions entirely (killing the most significant efficiency gain in a generation) or let them through blind (accepting unquantified risk from systems that can hallucinate, drift from intent, or be adversarially manipulated).

Both paths are bad. One stalls adoption. The other guarantees catastrophic failures that will trigger regulatory overreaction and set the entire space back years. The longer the gap between what agents can do and what can be verified about their reasoning, the worse the eventual correction will be.

Our approach: enable, don't gate

We're not building a wall between agents and commerce. We're building the instrumentation that makes the wall unnecessary.

Our thesis is simple: if you can measure the quality of an agent's economic reasoning in real time — before the authorization — then issuers can confidently approve transactions they would otherwise decline. The verification layer doesn't slow adoption. It accelerates it by giving institutions the evidence they need to say yes.

Concretely, that means:

Process-based evaluation (scoring how the agent reasons, not just what it decides). Six behavioral dimensions measured at the point of authorization. Scores that travel natively on existing card network rails. No new network, no agent-side SDK, sub-13ms added latency. Issuer-agnostic — works across any card program, any processor, any agent framework.

The goal isn't to judge agents. It's to make their reasoning auditable, governable, and transparent — so that every participant in the value chain (the human principal, the issuer, the network, the merchant) has the information they need to make trust decisions at machine speed.

What we're building toward

A world where agent commerce scales as fast as the technology allows, because the trust infrastructure scales with it. Where an issuer doesn't have to choose between innovation and risk. Where a human delegating spending authority to an agent has the same confidence in oversight that they have when they hand a corporate card to an employee — because the verification is continuous, real-time, and built into the rails.

We publish our research openly because this problem is too important and too urgent for any single company to solve in isolation. The frameworks need to be scrutinized, challenged, and adopted industry-wide. We're building a company, but we're also building a standard.

Founder