We built the internet on a simple assumption: the account holder is present, paying attention, and clicking "buy" themselves. Agentic commerce challenges that assumption fundamentally. When an AI agent books the flight, refills the prescription, or moves money across accounts, the human who authorized it may be asleep, offline, or three decisions removed from the transaction.
That shift is exciting. It is also where things can go seriously wrong. After nearly two decades building engineering and product systems, I keep coming back to one conviction: the best way to approach agentic commerce is to anchor every agent to a verified human. Call it a human trust harness. The agent can act, but a real, proven person remains accountable for what it does.
This is a sponsored write-up, in collaboration with our friends at Veriff. Learn more about Veriff’s trust infrastructure platform for identity, biometric, and business verification.
What makes this harder than it sounds is something I have written about before: trust comes from architecture, not from promises or process, but from the structural decisions we make about how systems are built, owned, and controlled. In agentic commerce, those decisions will determine whether the identity layer holds or becomes the weakest link.
Accountability: More Actors, More Layers, More Need for Proof
Traditional commerce has two parties and a relatively clear line of responsibility. Agentic commerce adds layers. You have the human, the agent acting on their behalf, the platform hosting the agent, and the merchant on the other end. Each layer is a place where accountability can slip.
When something goes wrong, someone has to answer for it. That only works if every action traces back to a verified human identity. As we explored in Who's verifying your verifier?, when responsibility passes through multiple unseen environments, attribution becomes harder and liability becomes blurred. Agentic commerce will significantly amplify that dynamic.
Guardrails: The Room for Error Is Much Wider
An agent can make a thousand decisions in the time a person makes one. That speed is the point. It also means the room for disaster is dramatically wider. Account takeover, which used to happen at human pace, can now compound at machine speed before anyone notices.
There are two main ways to ground an agent's identity, and the distinction matters. Cryptographically-signed identities are elegant, and the market will likely accept them more readily. But they carry a quiet assumption: the device is always in the right person's hands. That assumption fails the moment a device is lost, shared, or compromised.
Liveness-powered biometric authentication does not rely on that assumption. It confirms that a real, living person is present at the moments that count. It offers the right balance of security and convenience for high-risk agentic actions: strong enough to deter fraud, smooth enough that people actually use it.
Scalability: Trust Cannot Stop at Onboarding
Most identity systems do their meaningful work at sign-up and then carry that decision forward indefinitely. In agentic commerce, that model is likely to break.
An agent's behavior changes over time. Risk shifts by the hour. We cannot rely solely on the onboarding flow to sort out trust for every future action. What the industry needs is near-real-time, continuous risk assessment: the ability to re-verify at the moments that matter most, when a transaction is unusually large, unusually fast, or simply out of character.
This brings me back to architecture. As I outlined in The integrity of the verification pipeline, orchestration-based identity stacks that route verification through multiple third-party APIs may look capable on the surface, but each handoff strips context and diffuses accountability. The question for every platform will be: who owns the trust layer, and who is accountable when it fails?
What Payments Platforms and Fintechs Should Consider in the Next 12 Months
Three areas stand out:
Transparent auditability. As agents transact at machine scale, disputes and accountability questions could multiply. Clear, tamper-evident records of who authorized what are essential from day one, not as an afterthought.
Multi-provider standardization. Design for interoperable standards rather than dependence on a single agent provider, so the ecosystem can remain open, resilient and competitive.
A distinct trust layer. Identity, authorization and payment execution shouldn't become one undifferentiated system. Separation of concerns can provide an important architectural guardrail, with clear ownership and accountability across each layer.
Agentic commerce is coming whether the infrastructure is ready or not. The organizations that succeed will not simply be the fastest movers. They will be the ones who can demonstrate, at every step, that a real human stands behind the machine, and that the systems verifying that claim are built with genuine accountability, not just assembled from invisible parts.
