Visa Intelligent Commerce and Mastercard Agent Pay are the two card-network responses to agentic commerce, both launched in 2025 and both built on the same underlying design pattern: tokenise existing card credentials so that an AI agent can transact against the cardholder's existing payment relationship without holding any new instrument. Mastercard announced Agent Pay in April 2025; Visa announced Intelligent Commerce in October 2025. Neither product is on a tokenised rail in the sense the rest of this resource uses the term. Both run on existing card networks with their existing settlement, dispute, and fee mechanics intact, and neither contemplates the agent holding tokenised money or transacting in a tokenised asset. We track them because they are the legacy payment networks' positioning move for the agent-driven commerce wave, and because they set the comparison point against which the chain-native primitives (covered in x402) need to be read.
What each is
Mastercard Agent Pay was announced 29 April 2025 as the network's "agentic payments technology" with named launch partners including Microsoft and IBM, and a developer programme allowing agent builders to integrate Mastercard's tokenised credentials into agent workflows (Mastercard press release, 29 April 2025). The product extends Mastercard Token Service, the network's existing tokenisation infrastructure used for digital-wallet credentials, to a new credential class called Agentic Tokens. The agent presents the Agentic Token at checkout, the network resolves it to the underlying card account, and the transaction settles on the existing Mastercard rail. The cardholder remains in the legal seat as the principal; the agent acts under delegated authority documented at credential issuance.
Visa Intelligent Commerce was announced October 2025 with a similar architectural shape and a slightly broader product surface (Visa Intelligent Commerce overview). The named primitives are Visa Trusted Agent Protocol (the credential and authority framework), Visa Acceptance for AI Agents (the merchant-side integration kit), and Visa Tokenized Asset Platform Connect for the small subset of flows that touch tokenised assets. The launch partners include Anthropic, Microsoft, OpenAI, Mistral, Perplexity, Stripe, and Adyen on the integration side, with consumer-facing partners across travel and retail. Visa's framing places more weight on the trust-and-identity layer than Mastercard's launch positioning did, with explicit reference to verifiable credentials and the dispute-resolution apparatus the network has built up over decades of carded commerce.
The technical mechanics on both networks share the same shape. The cardholder signs up for the programme and grants authority to an agent (or to a category of agents). The network issues a tokenised credential bound to the cardholder's account but distinct from the primary account number. The agent presents the credential at checkout against any merchant accepting the network's standard rails. The network resolves the credential, applies any conditions the cardholder set at issuance (per-transaction caps, merchant-category limits, time windows), and authorises or declines through the standard flow. Settlement, clearing, and disputes run through the existing card-network plumbing.
How they differ from x402
The split with x402 is sharp on three axes. First, the rail. x402 settles on a chain in the asset the server names (typically USDC); the card-network programmes settle through Visa or Mastercard authorisation and clearing with the merchant receiving fiat after standard settlement timing (T+2 in most markets). The card programmes inherit the full card-rail cost stack: interchange, network assessments, and acquirer fees in the 1.5 to 3.5 percent range. x402 inherits the chain-fee stack, which on Base or Solana is materially smaller at typical agent transaction sizes.
Second, the holder model. x402 contemplates an agent holding a wallet and signing transactions against it. The card-network programmes contemplate an agent operating against the principal's existing card account without holding any tokens itself. Both are agentic in the sense Foundations Chapter IX uses the term (Part 1); the chain-native pattern is one of several shapes. The card-network pattern is operationally simpler in jurisdictions with mature card infrastructure because it requires no new licensing on the agent side and no new wallet infrastructure. The chain-native pattern is operationally simpler for cross-border and machine-to-machine settlement.
Third, the ecosystem. The card networks bring decades of acceptance infrastructure (tens of millions of merchants worldwide), a mature dispute apparatus, and an existing regulatory perimeter that handles consumer-protection questions. The chain-native pattern brings native composability with tokenised assets, programmable conditions enforceable at the contract level, and lower per-transaction costs at small ticket sizes. Neither dominates on every axis. For a tokenisation product team, the integration question is which inbound and outbound primitives the product needs to recognise, not which to bet on as the winner.
The awkward fit for both card programmes is the machine-to-machine settlement case where neither party is a merchant in the traditional sense. An agent paying an API for an inference call is not a card-present transaction in any meaningful way; the transaction sizes are typically well below the floor at which card-rail economics work. Both networks have positioned for the agent-to-merchant retail flow (an agent buying a flight, ordering supplies, completing a checkout a human would otherwise complete). Neither has a positioned answer for the agent-to-agent or agent-to-API flow at sub-dollar sizes, which is where x402 plus a stablecoin sits naturally.
What it tells us
Two structural reads. The legacy networks are taking the agentic wave seriously: both moved in 2025 with named launch partners on the agent-builder side (AI labs, major cloud platforms) and the merchant-acceptance side (major payment processors). Neither is doing a token strategy on the asset side; both are extending the credential-and-authority framework they already operate. The bet is that the agent-to-merchant retail flow will be where the volume lands and that the existing card rails are the right surface for it.
The strategic positioning differs. Visa's framing places more weight on the trust-and-identity layer and gestures toward tokenised-asset integration through Visa Tokenized Asset Platform Connect, hedging that the chain-native flow is a meaningful adjacent surface. Mastercard's framing is tighter on the card-rail extension and lighter on the tokenised-asset side. Neither has shipped a production integration bridging its agentic credential to a chain-native tokenised settlement asset; whether either ships such a bridge in 2026 is one of the open questions we track.
For an APAC tokenisation operator, the immediate operational implication is small. Card-rail agent payments on Visa or Mastercard work the same way in Singapore or Hong Kong as in the US, modulo local acquirer relationships. The structural implication is bigger: if the card networks capture the agent-to-merchant retail flow, the chain-native primitives end up living in the agent-to-agent and agent-to-API surface rather than the consumer-facing checkout. That would be a meaningful split with consequences for which tokenised assets see real agent-driven volume first.
Open questions
- Whether either network ships a production integration bridging its agentic credential to a chain-native tokenised settlement asset in 2026, and which asset is named first (USDC under Circle's regulated stablecoin pathway is the obvious candidate; bank-issued stablecoins or tokenised deposits would be more structurally interesting).
- Volume composition. Public reporting on agent-credential transaction counts and value through the first six months of each programme has not been disclosed at the cardholder or merchant level. We treat the absence of disclosure as a signal of small volume rather than a positioning choice.
- Dispute apparatus stress test. The card networks' existing dispute frameworks were built for human cardholders. Whether an agent acting outside its authority and triggering a chargeback creates novel dispute-resolution edge cases is undisclosed, and the first major dispute will likely set patterns.
- APAC merchant-acquirer roll-out. The launch partner lists for both programmes weight US and EU; the named APAC integrations have been thinner. Whether the local acquirers (NETS in Singapore, JCB in Japan, BC Card in Korea) integrate the agentic-credential programmes is the regional adoption question.
- Whether MAS or any other APAC regulator publishes guidance on the consumer-protection implications of agent-mediated card transactions, particularly around the chargeback path when the cardholder did not personally instruct the disputed payment.