The agentic payments space feels noisy today as multiple layers are evolving simultaneously.

At the coordination level, standards such as Model Context Protocol (MCP) and Agent-to-Agent (A2A) frameworks are defining how agents discover capabilities and collaborate across systems. These protocols matter because autonomous workflows rarely live inside a single application, with agents needing shared context and interoperability.

Above that, commerce itself is being restructured. Emerging efforts often described as Agentic Commerce Protocol (ACP) or Universal Commerce Protocol (UCP) are attempting to make buying machine-readable, allowing agents to exchange structured requests, pricing, availability and fulfilment terms directly between systems. In practice, this translates buying into machine-readable interaction, with structured quotes, inventory access and negotiation flows replacing traditional web interfaces.

Authorisation acts as a control layer across these interactions. Emerging models such as AP2 focus on mandates, encoding who granted authority, under what conditions and with what limits. Only once intent and permission are clear does the familiar payments layer appear: cards, bank transfers, network tokens, or programmable settlement rails.

The result resembles earlier phases of internet infrastructure, where communication, identity, and security standards developed independently before converging into a coherent system.

Traditional automation executes instructions defined in advance. However, agents operate within delegated boundaries, making decisions dynamically as conditions change. This changes the role payments play inside systems. Instead of discrete actions initiated by users, transactions increasingly emerge from continuous decision processes. We believe this will enable payments to reflect adjustments within ongoing commercial relationships, e.g. updating spend as usage, pricing, or performance changes rather than marking isolated purchasing decisions. A few examples include: A procurement agent may switch suppliers as prices move. A finance agent can rebalance subscriptions as usage evolves. A logistics system might release payment automatically once delivery conditions are verified. 

To support this shift, humans will need to define constraints rather than actions. Spending limits, policies and objectives replace individual approvals, and payments become the outcome of reasoning systems operating within authorised mandates. To what we alluded to above, rather than navigating interfaces, agents evaluate structured data such as pricing, delivery, usage and contractual terms, and transact when thresholds are met.

As intent moves earlier into policy and delegation, the centre of gravity in payment will start to shift with it. Risk management, compliance and reconciliation will move upstream, closer to decision-making rather than settlement. At the same time, responsibility becomes distributed across users, agent providers, software platforms, and payment systems, complicating traditional assumptions about who ultimately authorised a transaction.

As agents begin coordinating decisions across multiple systems, much of industry conversation has concentrated on payment execution and settlement, from stablecoins and programmable payment mechanisms to new initiation models such as HTTP-native payment triggers like x402. These approaches expand what becomes feasible, particularly for machine-to-machine payments involving APIs, data access or cross-border services. But settlement alone does not unlock agentic commerce. Agents can already move money through existing rails. The harder challenge lies in defining identity, responsibility and authority before funds move. As autonomy increases, rails become more interchangeable while governance becomes more complex, shifting innovation away from how money moves and towards how decisions to move it are validated.

Much of the early momentum around agentic payments has emerged in consumer commerce, where agents can operate within familiar transaction models. Companies such as Nekuda, Swap, and PayOS are building infrastructure for agent-led shopping experiences, while platforms like Crossmint are exploring agent wallets and programmable payment primitives that allow assistants to transact directly online. Alongside these efforts, card networks and payment providers are adapting credentials and tokenisation models to support transactions initiated by software rather than people.

The picture looks different inside businesses, where the B2B agentic payments landscape remains comparatively less crowded. While agents are increasingly embedded in finance and operational workflows, the systems that allow them to execute payments safely remain early. A small set of startups including Ralio, Natural, Payman, and Locus are beginning to tackle questions around agent identity, permissions and controlled execution. Around them sit adjacent platforms such as Zip, Vic.ai, and Routable, where agents increasingly shape spend management workflows, from procurement to accounts payable, while execution remains governed within individual financial systems rather than at the level of the agent itself. 

From our perspective, the opportunity spans several layers of the emerging payments stack.

We’re particularly interested in infrastructure that helps agents operate safely inside real economic environments: systems that make permissions programmable, decisions observable, and financial actions auditable. Equally compelling are tools that make commerce machine-readable - enabling businesses to expose pricing, availability and contractual logic directly to agents rather than interfaces.

We also expect new forms of orchestration to emerge as agents coordinate across vendors, services, and financial systems. As software starts to manage ongoing commercial relationships rather than isolated transactions, platforms that simplify how decisions translate into execution  may become critical connective tissue across the stack. 

As agent-driven commerce expands, financial infrastructure itself will begin to adapt to machine-driven activity. In cross-border environments, demand may increase for payment infrastructure offering programmable liquidity, faster settlement and API-native access to money movement. These developments do not define agentic commerce, but may become increasingly relevant as transaction volumes shift toward autonomous, machine-initiated activity, particularly where traditional rails struggle to support continuous cross-border transaction flows.