OpenAI just published a technical cookbook that walks developers through wiring up AI agents to make their own payments. The guide, titled “Controlled Agentic Commerce with AgentCore Payments,” pairs OpenAI’s Agents SDK with Amazon’s Bedrock AgentCore Payments service and a protocol called x402, which settles transactions in USDC on Base.
The practical upshot: AI agents can now autonomously pay for external resources, like data feeds or API calls, without a human clicking “approve” on every transaction. Instead, the application sets spending rules, and the agent operates within them.
How x402 actually works
The x402 protocol borrows its name from HTTP status code 402, “Payment Required.” That status code has existed since the early days of the web but was never widely adopted. It was essentially a placeholder for future use. Decades later, it finally has a job.
The flow works like this: an AI agent makes an HTTP request to an external service. Instead of getting the usual 200 OK response, it receives a 402 Payment Required response. That response includes payment terms, specifically the amount owed and the asset required, which in this case is USDC on Base.
The agent then checks whether the requested payment falls within its application-defined spending policies. If it does, the agent generates a cryptographic payment proof. That proof gets sent back to the service provider, the transaction settles on Base, and the agent receives the resource it was requesting. The whole loop completes with the kind of speed you’d expect from a layer-2 network: roughly 2 seconds to finality.
In OpenAI’s demo scenario, which models a supplier-research use case, each payment request costs 0.25 USDC.
The AWS piece of the puzzle
Amazon’s Bedrock AgentCore Payments service, which was first previewed in May 2026, handles the heavy lifting on the infrastructure side. It natively supports x402 challenges, meaning it can interpret those 402 responses and help agents generate the appropriate proofs within defined spending sessions.
Crucially, AgentCore Payments maintains audit trails for every transaction. The architecture addresses the “rogue agent” problem by keeping all spending approvals at the application layer. The agent never gets a blank check. It operates within guardrails that the developer sets in advance.
Why Base and USDC
The choice of Base as the settlement layer is deliberate. Coinbase’s layer-2 network offers low transaction fees, which is essential when you’re talking about micropayments measured in fractions of a dollar.
USDC, Circle’s dollar-pegged stablecoin, provides the price stability that makes this kind of automated commerce viable. An AI agent can’t really function in a payments role if the currency it’s using fluctuates 5% between the time it receives a price quote and the time it settles the transaction.
The x402 protocol was conceived from collaborative efforts by Coinbase, especially concerning the Base blockchain and USDC. Introduced as an open standard for internet payments, the x402 Foundation — which includes governance participation from Cloudflare — was established to support this vision, releasing version 2 in December 2025. Stripe subsequently integrated support for USDC payments on Base via x402 in February 2026.
What this means for developers and the broader market
OpenAI’s cookbook, released on August 13, 2026, isn’t just a technical demo. It’s a template. By publishing it as an open guide, OpenAI is effectively telling its developer community: this is how you should build payment-capable agents.
There’s also a competitive dimension worth watching. OpenAI partnering with AWS on payment infrastructure creates a powerful distribution channel. AWS has millions of enterprise customers already running workloads on its cloud. If AgentCore Payments becomes a standard feature of AWS-hosted AI applications, the x402 protocol could achieve the kind of adoption that competing agent-payment standards will struggle to match.
The flip side is concentration risk. If the dominant payment protocol for AI agents runs through two of the largest technology companies on Earth, the transactions settle on Base, which is decentralized in the network sense, but the agent infrastructure feeding those transactions is thoroughly centralized.