Beyond agentic payments: The operating layer for the Agentic Economy

"Are we behind on agentic payments?" is the question senior treasurers and heads of payments have started asking us. It is the right question to start with — but not where to stop.

Payments are the first visible use case of AI agents. But the larger shift we are seeing is software agents becoming economic actors. They will need to hold value, request authority, buy services, meter usage, settle transactions, create receipts, and be accountable when something breaks. That is bigger than payments. It is the operating layer for the Agentic Economy.

Agentic payments have moved from theory to working infrastructure. Six protocols, the four largest card networks, OpenAI, Anthropic, Google, and roughly every major Layer 1 (L1) blockchain are committed. USDC is settling 98.6% of agent-initiated transactions, per Keyrock's joint cohort study. This infrastructure is already being used across the ecosystem, and the settlement-asset convergence is the natural result of being the only programmable dollar with the regulatory clarity, reserve transparency, chain coverage, and operational track record people and companies trust.

However, volumes of agentic payments are hardly comparable with today’s fiat, or even stablecoin, transactions. Large-scale volumes are typically driven by large institutions, but for agentic payments, institutions are figuring out key constraints around identity, delegation, policy, audit, recourse, and transaction controls (to name a few). These are the bars treasurers have to clear before agent transactions move from ecosystem-experiments into institutional-scale volume. Therefore the strategic question we are discussing today with business and treasury leaders, is what to build in this planning window?

Agentic payments are in production across the ecosystem

As of writing, there are six standards that enable the agent payments layer: Anthropic's Model Context Protocol (MCP), OpenAI and Stripe's Agentic Commerce Protocol (ACP), Google's Universal Commerce Protocol (UCP) and Agent Payments Protocol (AP2), Coinbase and Cloudflare's x402, and Visa's Trusted Agent Protocol (TAP). Anthropic donated MCP to the Agentic AI Foundation last December, with OpenAI and Block as co-founders. All these actions speak to the overwhelming direction of the industry to open standards under neutral governance.

The card networks aren't waiting for demand to arrive either. JPMorgan Payments, Mastercard, Visa, American Express, PayPal, Coinbase, and Salesforce have all moved to position for agentic payments. The identity, tokenization, consent, and merchant-onboarding layer is being built ahead of broad demand, by the institutions that already own the consumer-side trust infrastructure.

Agentic payments today: real volume vs the forecasts

Individuals and institutions that have tried agent payments come away knowing the technology works. But institutional volume hasn't shown up yet.

For agentic payments today, MPPScan, the public dashboard that tracks real, on-chain agent payments on Machine Payments Protocol (MPP), shows about $240,000 in all-time volume across 1.7 million transactions, spread over roughly 144,000 agents (as of July 17, 2026). CoinDesk's March read on x402 found around $28,000 in daily real volume, much of it from ecosystem testing and a single team's API wrappers. And Keyrock's year-long study counted 176 million agent payments, most under $0.30, adding up to just $73 million.

But, future forecasts of agentic payments run much larger. Bain forecasts 15-25% of US e-commerce mediated by agents by 2030. BCG identifies $200 billion in net-new technology services demand just to integrate agents into existing enterprise software. Even corporate treasury, slower to move, is starting: the Strategic Treasurer survey found 22% of companies using agentic AI in some form, with payments-initiating activity at a small fraction of that.

So obviously the projections describe a market that hasn't arrived. The rails work and are being used at scale across the ecosystem. But the institutional infrastructure to authorize, govern, audit, and underwrite an agent payment at institutional volume is still being created. Until the “operating layer” is built out, volume will struggle to grow.

How to build the operating layer for the agentic economy?

While USDC has emerged as the settlement asset of the agentic economy, the constraints for the rest of the operating layer sit in the following areas: identity, delegation, policy, audit, recourse, and transaction controls. These elements form three core parts of the agentic economy operating layer.

The first piece is identity, delegation, and recourse. The Bank for International Settlements (BIS) and the International Monetary Fund (IMF) have both flagged these as the unresolved layer. Visa TAP and Mastercard Agent Pay are early attempts at solving it; so is the Agentic Risk Standard paper. None has yet converged into a workable institutional standard, but most proposals share the same primitives. For now, no enterprise treasury team will authorize agent-initiated payments at institutional volume until "who's allowed to act, with what limits, against whose balance sheet, and with what recourse if it breaks" has a clean answer.

The second is policy and audit. Pay-per-API and pay-per-inference payments are often worth a fraction of a cent, so they only work when settling one costs even less. Nanopayments (sub-cent, instantly verified, bundled into single onchain transactions) is the unlock, and it's shipping into production now. Audit trails and policy controls have to be designed from the start (e.g. per-transaction and daily limits, approved-counterparty lists) which is often the highest friction point to getting an AI project started.

The third is transaction controls. Most organizations have not yet defined how an agent gets authorized to transact, within what limits, under which scheme and network rules, and how disputes resolve when the buyer is an agent. The early movers are working this out inside controlled pilots, well before opening it to live volume.

The complete institutional layer wrapped around the agentic economy (digital assets, custody/wallets, identity, auditability, settlement finality) has to clear two bars at once: it has to be end-to-end, and it has to be regulated. Together they make the layer deployable inside a regulated institution, and that combination is rare in the market today.

What we're hearing from senior treasurers and heads of payments is that the elements of the operating layer for the agentic economy are no longer a future problem. It's the work of the next two to four quarters.

What to prioritize for the Agentic Economy

From our conversations with institutional treasurers and heads of payments, three priorities define the operating-layer build over the next four quarters.

1. Go beyond agentic payments, to the agentic economy, for institutional scale. The goal is not just to make an agent payment work; it is to make agent-led economic activity governable, auditable, repeatable, and safe enough for institutions to scale effectively.  
2. Document the agent-policy framework before it's needed. The institutions that have "what an agent can do, within what limits, with what monitoring, and which department owns the loss" written down in 12 months are the ones that can scale when institutional volume arrives.
3. Start planning across the full operating layer, today. Stablecoins are the most visible decision, yet the infrastructure and processes underneath also need to be chosen:  blockchains, wallets, identity, audit, and settlement-finality. The institutions planning today are choosing partners who can pass both the end-to-end architecture AND regulatory requirements to help with the heaviest lift — documenting the processes and guardrails for AI to follow.

The agentic economy will be bigger than today's numbers suggest, extending beyond payments into treasury, capital markets, programmable value exchange, and machine-to-machine coordination operating inside regulatory boundaries. The honest answer to "are we behind on agentic payments?" is no, but the bigger question you should be asking is “are we behind on the agentic economy?” The answer there is most likely yes. 

‍