Executive Summary

In the six months between September 2025 and March 2026, every major player in global payments made a significant move. OpenAI and Stripe jointly launched the Agentic Commerce Protocol. Google announced the Universal Commerce Protocol to a coalition of over 30 retail and fintech partners. Visa and Mastercard each released agent-specific payment frameworks. Coinbase's x402 protocol cleared more than 15 million transactions on Base. Stripe and Tempo co-authored the Machine Payments Protocol and submitted it for IETF standardisation.

The timing is not a coincidence. It reflects a shared recognition that the payments infrastructure built over the past three decades was designed for humans: it assumes someone is sitting at a browser, filling out forms, clicking buttons, and completing verification steps. AI agents do none of those things. They need programmatic interfaces, near-instant authorisation, and settlement infrastructure capable of handling transactions at fractions of a cent. The existing stack was not built for this, and everyone in the industry knows it.

What is emerging in response is a two-layer architecture. The upper layer handles commercial orchestration, how an agent identifies where to buy and initiates a transaction. The lower layer handles settlement, how money actually moves. These two layers will evolve on separate tracks, driven by different players with different incentives. Neither is sufficient without the other, but they should not be conflated.

Part I: Commercial Orchestration

The orchestration layer determines how a transaction comes together: service discovery, session management, and the handoff to payments. Two distinct categories of use case have emerged, and they differ enough in structure that treating them as one market would be a mistake.

1.1 Agents Acting on Behalf of Consumers

The immediate obstacle here is not payment mechanics, it is access. E-commerce platforms were built for human navigation. An agent’s native navigation process shouldn’t be scrolling through a product page, interpreting promotional banners, or clicking an "add to cart" button. Before any payment can happen, merchants should offer structured, machine-readable endpoints. Most do not yet have them.

ACP — Agentic Commerce Protocol (OpenAI and Stripe)

OpenAI and Stripe launched ACP in September 2025. The protocol is built around secure payment delegation at the moment of purchase: a user's payment method is saved in ChatGPT, and when they confirm a purchase, Stripe issues a Shared Payment Token (SPT), a single-use credential scoped to that specific merchant and cart total, which is passed to the merchant via API. The merchant retains full Merchant of Record status and processes payment through their existing stack. Stripe's SPT is, as of this writing, the first live implementation of this delegation model, and is compatible with OpenAI's Delegated Payment Spec. Importantly, other PSPs can implement their own Delegated Payment Spec integrations, making ACP genuinely open at the payment layer.

ChatGPT Instant Checkout launched in September 2025 for U.S. users and was shut down in March 2026 after near-zero purchase conversion. OpenAI has since pivoted to discovery: ChatGPT surfaces products and redirects users to merchant sites and apps for checkout. ACP the protocol survives in narrower form, powering dedicated in-ChatGPT apps for a small number of large retailers. Merchants must apply to participate, and OpenAI controls which are shown and in what order.

UCP — Universal Commerce Protocol (Google)

Google's entry came at the NRF Conference on January 11, 2026, announced by Sundar Pichai. UCP was co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, and endorsed by over 20 additional partners including Adyen, American Express, Best Buy, Mastercard, Visa, Stripe, and The Home Depot. The protocol is explicitly designed to be compatible with Google's own Agent Payments Protocol (AP2), the Agent2Agent (A2A) standard, and the Model Context Protocol (MCP), a deliberate "high ground" interoperability play.

UCP works through a capability manifest called a UCP profile: merchants publish a structured JSON document at /.well-known/ucp under their domain, declaring what transport methods, checkout capabilities, and payment handlers they support. Agents read this directly, without any intermediaries. Google Pay serves as the default payment instrument for UCP-powered checkout, with PayPal announced as an upcoming option.

This architecture reflects a deliberate strategic choice. Google has no need to broker transactions, an approach that would expose it to margin pressure, liability, and regulatory scrutiny. What it requires instead is comprehensive indexing of the commerce web. UCP positions Gemini as the primary discovery layer for agent-driven shopping, fully leveraging Google's established search expertise. The profile model keeps Google upstream: indispensable for discovery, yet invisible at settlement.

ACP and UCP are not competing for the same position. ACP operates as a controlled, curated environment, OpenAI acts as gatekeeper, merchants must apply, and the experience is optimised end-to-end within ChatGPT. UCP functions as an open catalogue: merchants self-publish their profiles, any compatible agent can read them, and Google controls the discovery surface without controlling the transaction. Lower onboarding friction, broader potential reach, but less merchant hand-holding.

1.2 Agents Transacting with Other Agents

The second category is structurally different. When both parties of a transaction are autonomous agents and no human merchant is in the loop, the familiar trust anchors disappear. There is no terms-of-service agreement, no consumer protection statute, no credit card chargeback right. The question is how two systems that have never interacted can exchange value without either needing to trust the other.

ERC-8183 and ERC-8004 (by Ethereum Foundation and Virtuals Protocol)

This is the problem ERC-8183 was designed to address. Proposed on March 10, 2026 by the Ethereum Foundation's dAI team alongside Virtuals Protocol, the standard structures each transaction as a three-party job: a Client that commissions work, a Provider that performs it, and an Evaluator that certifies completion. Payment is held in escrow by a smart contract and released only when the Evaluator signs off. Neither the Client nor the Provider needs to assess the other's trustworthiness, the contract enforces the outcome mechanically.

ERC-8004 provides the identity layer underneath. Agents register on-chain and build a reputation score from their transaction history, creating a credibility signal that persists across interactions. The protocol design is sound; the constraint is bootstrapping adoption at scale.

In practice, the existing ecosystem activity is concentrated within the Virtuals Protocol platform, where an orchestrator agent called Butler breaks complex tasks into sub-jobs and routes them to specialist agents across the network. The broader developer community has not yet engaged at scale. ERC-8183 represents the attempt to make this pattern open and permissionless.

One structural point is worth making here: the two tracks imply different settlement requirements. E-commerce can run on card rails because human buyers are in the loop. Pure agent-to-agent commerce would likely require stablecoin settlement, as the economics of card fees may not work if services are transacted at small sizes and high frequencies.

Part II: Settlement

If the orchestration layer determines what is being transacted, the settlement layer determines whether the transaction actually closes. Five protocols are currently competing in this space, and they differ in ways that matter.

2.1 Delegated Payment Spec / SPT (Stripe)

Stripe’s approach extends its existing card infrastructure rather than replacing it. When a customer authorizes an agent, Stripe provisions a Shared Payment Token (SPT) that the agent stores. At transaction time, the agent presents this time-limited, amount-capped token to the merchant. Settlement runs through Stripe’s standard card processing stack.

On the back end, Stripe has built bridges to Visa Intelligent Commerce and Mastercard Agent Pay (which issues agentic network tokens), so merchants see a single integration surface regardless of which card network sits underneath.

This architecture is well-suited for standard retail purchases and a broad range of transactions, including high-value agent-to-agent payments where consumer protections such as chargebacks are advantageous. However, it is less suitable for high-frequency, micro-value agent-to-agent or machine-to-machine payments, where transaction amounts are typically fractions of a cent and volumes can reach thousands of transactions per minute.

2.2 Visa Intelligent Commerce and Mastercard Agentic Token

Both networks have re-engineered their tokenisation infrastructure to accommodate agent-driven transactions. Real card numbers are replaced with dynamic encrypted tokens that carry metadata about the authorising agent: its identity, spend limit, validity window, and the set of permitted merchants. Settlement remains on the existing card network, no new rails required.

Both networks have moved beyond pilots. Mastercard processed the first fully identified agent transaction in September 2025, in partnership with Commonwealth Bank in Australia. Visa completed initial deployments across European markets through its Agentic Ready programme. The infrastructure is proving capable, but the fee floor creates a genuine constraint, neither network can serve the sub-dollar micro-payments that agentic commerce may eventually require at scale.

2.3 x402 (Coinbase)

x402 takes a fundamentally different approach. Rather than adapting card infrastructure, it builds on HTTP itself, specifically the 402 "Payment Required" status code, which has been part of the HTTP specification since 1997 but was never widely implemented. When an agent requests a resource that requires payment, the server returns a 402 with payment parameters attached. The agent signs an authorisation, and a Facilitator completes atomic on-chain settlement in USDC or other supported tokens, typically within about two seconds. No account setup, no API keys, no KYC required. The x402 Foundation, established by Coinbase and Cloudflare, governs the open standard.

By end of 2025 the protocol had processed over 100 million transactions across Base, Solana, and Polygon. However, this number requires significant qualification: analysts at Artemis, writing in February 2026, estimated that a substantial portion of x402 activity reflects self-dealing and infrastructure testing rather than genuine commerce. The protocol's annualised payment volume is reported at around $600 million, but concentration and quality-of-volume concerns are real and material.

There is no structural fee floor preventing micropayments, x402 was purpose-built for sub-cent transactions. The current gap is adoption depth and real-commerce density, not architecture.

2.4 Nanopayments (Circle)

Circle's Nanopayments protocol is explicitly x402-compatible, it uses the HTTP 402 trigger per the x402 standard, and builds a batched settlement architecture on top. Rather than settling each transaction on-chain individually, it uses a pre-funded Circle Gateway account on the buyer side and EIP-3009 off-chain signatures for each payment, with periodic batched settlement to the blockchain. Gas costs are absorbed once at deposit rather than per transaction, which makes payments as small as $0.000001 economically viable.

The trade-off is that both parties must deposit into Circle Gateway first, which limits the protocol to a semi-closed network in its current form. Nanopayments launched on testnet in March 2026, across 12 supported chains. The fee structure is compelling for high-frequency micro-payment use cases if Circle can solve the onboarding friction at scale.

2.5 MPP — Machine Payments Protocol (Tempo and Stripe)

MPP is the most architecturally ambitious of the five. Co-authored by Tempo, a payments-focused Layer-1 blockchain built by Stripe and Paradigm, and Stripe, MPP uses the HTTP 402 trigger and allows merchants and agents to select from multiple settlement rails within the same framework. A developer does not need to choose between stablecoin and fiat infrastructure at build time; the agent can make that decision at runtime based on transaction requirements. Available options include:

• Tempo: stablecoin in both per-transaction on-chain and off-chain batched modes

• Stripe: payments using Shared Payment Tokens (SPTs)

• Card: payments using encrypted, single use network payment tokens and dynamic data provided by a card network for machine-initiated transactions.

• Lightspark: Bitcoin Lightning payments

MPP introduces a "session" primitive, an OAuth-like model where an agent authorises once and pre-funds an account, enabling real-time automated settlement across subsequent interactions without an on-chain transaction per payment. The core spec has been submitted to the IETF for standardisation as the official implementation of HTTP 402. At launch, the Payment Directory includes integrations with over 100 services. MPP launched on mainnet on March 18, 2026.

Stripe's dual role here is worth noting. It co-authored the protocol and is simultaneously one of the payment options within it, which means it captures value whether developers adopt MPP for its flexibility or specifically for its card capabilities.

Part III: Where Things Stand

3.1 State of Play

Every major protocol covered in this report launched within the past six months, but commercial activity has not followed. On the settlement side, x402 leads in transaction count but real daily commerce volume runs at approximately $28,000. On the orchestration side, ACP's flagship checkout product was shut down after near-zero conversion rates. Newer protocols like ERC-8183 and MPP face the same pattern: narrative running well ahead of actual deployment. The market has reached a distinct inflection point, the protocol architecture is largely built, but scaled commercial application has not yet begun.

The central challenge is severe fragmentation at the orchestration layer. Merchants must navigate multiple independent standards, each with its own SDK, authentication flow, and compliance requirements, a situation that makes integration costs unpredictable and discourages participation. Historically, this degree of fragmentation resolves through an aggregation layer that unifies access above the competing standards. This time may be different. Every platform with meaningful agent traffic, OpenAI, Google, or Microsoft, has a direct incentive to build and maintain its own closed surface rather than route users elsewhere. This logic is playing out simultaneously across markets: China, Southeast Asia, Korea, Japan are each developing regional closed-loop ecosystems anchored by dominant super-apps. The more likely outcome is parallel regional closed ecosystems, not a single open standard. The aggregation layer that merchants need is more likely to be built by third-party infrastructure providers serving merchants than by the platforms competing for agents, the incentives simply do not align.

3.2 Where the Opportunities Are

Two distinct opportunities follow from this analysis.

Settlement Infrastructure: The Most Certain Near-Term Opportunity

Regardless of how fragmented the orchestration layer becomes, every agent has the same underlying problem: payment. A clear asymmetry is emerging — orchestration fragments by platform incentive while settlement consolidates by developer necessity. A developer cannot maintain separate payment integrations for every orchestration surface their agent might operate on. As the number of platforms grows, the economic pressure toward a single unified payment integration only intensifies.

This sets a concrete product requirement for agent wallets: multi-rail support is not optional. Card rails — SPT, Visa Agentic Token, Mastercard Agentic Token — cover traditional merchant commerce. Stablecoin rails — x402, MPP session payments — cover on-chain API services and agent-to-agent transactions. Both use cases are real today and will not merge into a single rail in the near term. The flexibility requirement sits on the agent side, not the merchant side: merchants choose which payment rails they support, a relatively stable decision within their control. Enterprises provision their agents with stablecoins and a delegated card; the agent pays using whichever rail the counterparty accepts. A wallet that handles both rails within a single integration is what makes a general-purpose agent viable across the full range of counterparties it will encounter. That value compounds across every transaction, every platform, and every ecosystem, building infrastructure depth that is difficult to displace.

Agent-to-Agent Commerce: The Underdeveloped Layer

The real gap is at the application layer. The A2A economy today is largely confined to crypto-native use cases, agents querying on-chain data, interacting with DeFi protocols, and processing blockchain transactions. The market has not moved beyond crypto infrastructure. Technically, having agents commission real-world services — data analysis, content production, legal research, code review — is already possible at the protocol level. The developers building those services and structuring them as agent-payable, per-call APIs largely do not exist yet. That is the actual gap, and it is the least competed direction in the space.

This opportunity is currently constrained by a genuine cold-start problem. Identity systems like ERC-8004 require transaction density to generate meaningful trust signals — an agent with no transaction history has no credibility, which limits who will transact with it, which limits how it builds history. Microsoft has projected approximately 1.3 billion active AI agents by 2028; the current population is orders of magnitude smaller. This is not a gap that resolves automatically, it is the specific barrier that keeps competition low, and crossing it first is the position worth taking.

The deeper implication is a business model restructuring that extends well beyond payments. The internet's two dominant models, advertising and subscriptions, both assume the buyer is human. Agents are not influenced by advertising and do not need monthly access; they pay for the output of a single call. HTTP 402 per-call pricing offers API providers a fundamentally different structure: sell results rather than access, charge heavy users for what they actually consume, and stop subsidizing light users or over-provisioning for rare peak demand. Whether A2A commerce expands beyond crypto and whether HTTP 402 becomes a general pricing primitive for software are the same question asked from two directions.

Conclusion

Agent commerce will develop along two tracks that should not be conflated. Consumer-side agent commerce, agents purchasing goods on behalf of humans, will run on card rails, and its pace is determined by enterprise authorization frameworks and consumer trust in new payment surfaces. Agent-to-agent commerce, software paying software for APIs, compute, and data, is technically ready on stablecoin rails, waiting for the agents and services that need it to be deployed at scale.

The end state will be a two-layer stack evolving in parallel: the orchestration layer governing how transactions are initiated, and the settlement layer governing how value moves.

For builders, the priority is breadth of integration. Infrastructure that can route any agent transaction through whichever protocol the counterparty requires, without surfacing that complexity to the application layer, will hold a structurally strong position when the market scales. This is not a visible product category, but it is the kind of infrastructure that compounds quietly and becomes difficult to displace.

The trigger for commercial scale is not better protocols. It is the moment enterprises delegate spending authority to agents: auditable transaction trails, budget authorization mechanisms, and clear liability when an agent makes a wrong purchase. When that moment arrives, the two most critical and currently unoccupied infrastructure positions are a multi-rail agent wallet covering both stablecoin and card within a single integration, and an accessible per-call service directory that lets developers with no crypto background expose their APIs to agent buyers. Both are open today. Both become essential at the same moment: when spending agents deploy at scale, they will need to pay across rails and discover services to buy. Neither position works without the other.

About Foresight Ventures

Foresight Ventures is a top-tier crypto venture firm and one of the five most active investors globally in 2024. With our team in US and Asia, we are the first and only crypto VC truly bridging East and West. Our approach is research-driven and founder-focused: we back the infrastructure powering next-gen global payments — from stablecoins and on/off-ramps to real-world assets, while amplifying our portfolio through a strong media network. Our 150+ investments include Story, TON, Aptos, Morph, 0G Labs, Sentient AI, The Block, Foresight News, and many more.

For more information, please visit: Website | Twitter | LinkedIn