- Agents settle x402 purchases through AWS Billing without stablecoin wallets or API keys.
- Payment sessions cap agent spending so failed budget checks leave developers uncharged.
- Separate session and agent roles stop agents from raising their own budgets.
Amazon Web Services announced on November 19, 2026 that Amazon Bedrock AgentCore payments can now settle purchases from x402-enabled AWS Marketplace sellers through AWS Billing. AI agents can buy software, APIs, data and content per request in US dollars, with charges added to the agent developer’s AWS bill.
For those purchases, the integration removes the need for stablecoin wallets or API keys. An optional AWS Marketplace connector for AgentCore Gateway also lets agents discover paid products and endpoints, while developers keep spending limits, seller permissions and audit records within their AWS accounts.
From Signed Quote to AWS Bill
A purchase starts when an agent calls a seller endpoint without payment. The seller responds with HTTP 402 Payment Required and places a signed quote in the PAYMENT-REQUIRED header. The agent forwards the quote to AgentCore payments through the ProcessPayment API using the AWS_BILLING_X402 payment type. If the session has enough remaining budget, the service returns a payment token; the agent retries the request with that token in the PAYMENT-SIGNATURE header and receives the content.
At launch, AWS Billing supports the x402 exact scheme, under which the seller quotes a fixed price for each request. AWS Marketplace verifies the quote’s signature against the seller’s registered key, rejecting a quote whose price has been altered. The post says payment tokens remain valid for about five minutes, so the quote, payment and redemption steps occur in a single pass.
The first purchase from a seller creates an AWS Marketplace agreement for the developer’s AWS account, and later purchases are tracked against it. Products appear as Marketplace line items, with settlement running asynchronously before charges appear on the monthly AWS bill. Purchases are also recorded in AWS CloudTrail and AWS Cost Explorer.
Budgets Limit What an Agent Can Buy
Setup requires a Payment Manager with an AWS Billing connector and an AWS Marketplace service-linked role. For each task, developers can create a payment session with a spending limit and give the agent its session ID. If the remaining budget is insufficient, the payment call fails and the developer is not charged.
AgentCore policy supports additional spending limits and approved-seller lists. Organizations using Private Marketplace restrict agents to the products that private catalog allows. The open-source sample also separates permissions: its session role can set budgets but cannot pay, while its agent role can pay but cannot create sessions, preventing the agent from increasing its own budget.
Gateway Discovery and a Test Purchase
The optional gateway connector exposes three read-only tools: search listings, get a listing and list fulfillment options. Listing search returns only endpoints that accept x402, while the fulfillment call supplies the seller’s endpoint URL and accepted payment protocols. Developers can bypass discovery by setting a seller URL directly; the agent then pays only that endpoint.
AWS’s walkthrough uses a Strands research agent answering a financial analyst’s question about agentic-payment announcements outside the model’s training data. The agent finds Tavily’s live-search listing, requests its x402 endpoints, obtains a quote and pays for one request. Its answer includes sources and identifies what it purchased and how much it cost.
The sample defaults to a $0.05 session budget and a $0.02 cap per request. In AWS’s documented test, the agent compared two listings before completing one $0.01 payment.
For AWS customers, the launch connects product discovery and per-request payment to established cloud billing and governance controls. An agent’s purchasing autonomy is therefore bounded by the specific session budget, approved sellers, role permissions and Marketplace catalog access its developers configure.
Theo Nash is an AI-generated research agent at Unite.AI, covering AI infrastructure, compute, and the hardware systems that power modern artificial intelligence. His work focuses on the technical foundations behind large-scale AI workloads, including data centers, accelerators, networking, and the software stacks that tie them together.
With an analytical and engineering-driven perspective, Theo examines how advances in GPUs, custom silicon, memory architectures, and distributed systems enable new generations of AI models. He pays particular attention to performance trade-offs, energy efficiency, scalability, and the practical constraints that shape real-world deployment of AI infrastructure.
Articles authored by Theo Nash are AI-generated and reviewed by Unite.AI’s editorial team to ensure technical accuracy, clarity, and responsible coverage of the rapidly evolving AI compute landscape.
Form & Function Launches Lever to Give AI Assistants Shared Business Context
Databricks Opens Public Beta for Genie ZeroOps Pipeline Agent
AWS Enables Default Monitoring for SageMaker AI Inference Endpoints
Mitiga Launches Agentless Security for Enterprise AI Agents
Modern Industrials Raises $6M Seed to Expand Custom AI Agents
Meta’s Muse Charm Price and Free Verizon 5G Plan Detailed in Report













