AWS, Visa, Mastercard and Coinbase expand agentic AI payments

Visa’s partnership with OpenAI injects its Visa Intelligent Commerce protocol into ChatGPT‑style models, letting AI agents initiate tokenized card payments directly from large language models [1]. At the same time, Mastercard unveiled “Agent Pay for Machines” with 30 partners—including Stripe, Solana and Adyen—to let software transact with software, a move aimed at high‑volume, low‑value machine‑to‑machine commerce [1]. AWS added its own piece to the puzzle on May 7, debuting payment features for Amazon Bedrock’s AgentCore service that enable agents to buy data feeds, cloud tools and, eventually, consumer items like flight tickets [2].

The Bedrock toolkit relies on a novel “x402” protocol that uses the HTTP 402 status code to convey purchase instructions, then returns a proof of purchase to the agent. Payments can be funded from digital wallets, with Coinbase providing the x402 technology and Stripe’s Privy subsidiary handling fiat and stablecoin transactions [2]. Developers can cap spending per session and monitor activity through built‑in observability, giving enterprises control over autonomous spend. AWS rolled out the features alongside a major cloud agreement with U.S. Bancorp, which will migrate hundreds of banking workloads to the platform [2].

Fintech firms are racing to be “agent‑agnostic.” Coinbase announced a “Coinbase for Agents” offering that lets AI agents manage portfolios, improve cash efficiency and execute data‑driven trades, mirroring Robinhood’s approach [1]. The companies argue that customers will bring their preferred agents to these platforms, bypassing traditional bank apps. Analysts warn that this could erode banks’ direct relationships with consumers, as agents become the primary interface for payments and finance [1].

The competing rails—Visa and Mastercard’s tokenized card credentials versus Coinbase’s stablecoin‑based internet protocols—highlight a deeper contest over who will own the next generation of payment settlement [3]. As AI agents become capable of autonomous spending, the question for banks and retailers is whether to embed large language models into their own ecosystems or risk becoming a background ledger as customers transact through external agents [1]. The outcome will shape how money moves in an increasingly agent‑first digital economy.