The hottest debate in agentic commerce today may be whether stablecoins or fiat becomes the predominant payments paradigm, but the best companies actually building in the space are focused on something more subtle, and more meaningful – the agentification of payments.
This encompasses all of the complexity that surrounds the core transaction, and how to adapt it for the agentic economy. Things like the authorization logic, identity layer, reconciliation infrastructure, and dispute and fraud frameworks. In a world that was constructed around humans initiating and managing payments, this broader ecosystem needs to be reimagined.
This idea crystallized at a recent Go Restive panel on agentic payments. We invited leaders from Stripe, Coinbase, and Natural AI to speak about the current state of agentic payments, and share how they are actively shaping the next wave of commerce. Despite their different architectures, business models, and views on which rails will win, none of them are primarily focused on the core question of payment rails. The real innovation (and the real opportunity) is in agentification: the protocols, ledgering tools, and authorization logic that make it possible for agents to transact at all. That’s where each of these companies is spending its time, and where any startup entering this space should be looking. (And if you want to view that conversation in its entirety, do so here)!
Stripe is not trying to replace the card. In fact, they appear to be splitting the difference on the rail question, enabling both card and stablecoin payments. But their primary focus is the developer infrastructure that makes agents able to transact intelligently: the APIs and tooling that let an agent make a payment decision, manage multi-step purchasing flows, and handle failures without needing to understand the mechanics of what is happening underneath. The bet is that whoever controls the abstraction layer through which agents access payments infrastructure captures significant value, regardless of which rail that infrastructure sits on.
Coinbase's x402 is a protocol for machine-to-machine payment communication, not primarily a consumer crypto product. It uses the HTTP 402 status code, long reserved for payment-required responses, to create a standard for how an agent requests access to a resource and how a provider responds with payment terms. The settlement layer uses a crypto wallet, which is a great use case for microtransaction unit economics. But the more significant work is in the protocol itself: defining how agents negotiate, authorize, and confirm payments without human intermediation in a way that scales across the open web. They are actively building out that infrastructure, and creating the ecosystem to drive the agentification of all of the actions surrounding the protocol.
Natural AI's core payments flow still runs on ACH and traditional payments. Their work is in the messy management of those payments; things like permissioning, ledgering, and governance, in a B2B payments context.
Our panel could have ended up as a debate about rails and protocols (and we did get into that!). But we settled in a much more interesting place, around the fact that the enabling infrastructure is what needs to be built for agentic commerce to really start to take off.
The problems that need to be solved in the agentification layer are specific, technically demanding, and represent a large shift from a human-centric worldview of payments. These will become very important as we see the sector start to achieve scale. These could be as standards, as external startups, or as unique IP that companies build into their product stacks.
Permissioning. Which agent is authorized to spend what, under what conditions, with what approval logic, and with what ability to override? This is a dynamic, context-sensitive permissions layer that does not exist in any current system designed for agentic commerce.
Reconciliation and ledgering. When agents are executing thousands of transactions across multiple vendors, systems, and contexts, how does a finance team understand what happened and why?
Agent identity and authentication. Commerce requires an answer to the question of who is responsible when something goes wrong. Agents do not have persistent identity or accountability structures that translate into existing payments frameworks. Building authentication for agents in a form that satisfies merchants, processors, and regulators simultaneously is a foundational unsolved problem.
Merchant trustworthiness. Agents lack the ambient signals humans use to assess merchant legitimacy — checking reviews, reading the room, evaluating a site’s credibility. Merchant verification becomes a first-class technical problem when anyone can spin up 100 storefronts targeting agents overnight.
Disputes and error handling. What happens when an agent pays for something the principal did not want? The dispute infrastructure that exists today was designed around human behavior and human error. Agent-native failure modes will be different, and the resolution workflows will need to be rebuilt to match.
Fraud. Agents will introduce a new vector for fraud, whether through new forms of account takeover, manipulation, or external attacks. Tools for surfacing and resolving this need to be created.
None of this is to say the rail question is irrelevant. Existing platforms face a real architectural mismatch, and they are subpar for many use cases: B2B payments, agent-to-agent commerce, and direct content monetization are a few that seem ripe for a major platform shift. But the largest opportunity in agentic commerce is not splitting up the pie among existing payment players. It is in creating new categories of commerce, the enabling infrastructure, and the tools required for scale that represent the greatest opportunity.
The most credible signal in any emerging market is where the best builders are spending their time. In agentic commerce, that answer is unambiguous: on the enabling infrastructure that makes agentic commerce possible in the first place. The map is bigger than the debate, and most of it is still blank. If you are building in this space, we would like to hear from you!
© 2025 Restive®, Inc.
