{
  "id": 12931107,
  "title": "Pay-per-inference for AI agents: How BlockRun and Incarna use Amazon Bedrock AgentCore payments",
  "url": "https://urgent.news/2026/10/08/pay-per-inference-for-ai-agents-how-blockrun-and-incarna-use-amazon",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-08T18:33:29.000Z",
  "source": {
    "name": "AWS Machine Learning",
    "slug": "aws-machine-learning",
    "url": "https://aws.amazon.com/blogs/machine-learning/pay-per-inference-for-ai-agents-how-blockrun-and-incarna-use-amazon-bedrock-agentcore-payments/"
  },
  "original_language": "en",
  "account": "When an AI agent needs to run, it often requires purchasing something to complete a task, such as an inference, API response, web content access, or a call to another agent. These purchases are typically small and frequent, often amounting to a fraction of a cent each. They occur within the agent's loop without any human approval. Amazon Bedrock AgentCore payments offers a managed solution for these on-demand payments, with spending limits enforced by the infrastructure rather than the model.\n\nIn this post, we examine how Incarna implemented AgentCore payments to enable its agents to pay BlockRun for model inference on a request-by-request basis. BlockRun is a pay-as-you-go inference router that supports over 90 models from more than 15 providers, with each call independently quoted and settled. By utilizing AgentCore payments, the Incarna team streamlined the x402 payment support integration process from months to just days and successfully deployed an end-to-end pay-per-inference flow in production.\n\nThe primary challenge in paying for inference by the request is that it involves high-frequency, low-value transactions. An agent might execute hundreds of small purchases within a single session, each being a fraction of a cent. Traditional card networks are not equipped to handle such low-value payments. Building custom payment rails entails addressing multiple complex problems simultaneously, including determining where to hold funds, signing each payment, accommodating emerging payment protocols like x402, and preventing overspending by autonomous agents.\n\nAmazon Bedrock AgentCore payments addresses these challenges, providing a platform for building, connecting, and optimizing agents at scale, regardless of the framework or model used. AgentCore payments is a managed capability within Amazon Bedrock AgentCore that simplifies the payment integration process. It handles the payment protocol, connects to a wallet, signs the transaction, and enforces spending limits. Builders only need to integrate through a single managed service instead of assembling the components independently.\n\nAgentCore payments offers several advantages for pay-per-inference use cases. It supports managed wallets, enabling agents to provision their wallets using Coinbase CDP connectors, with the customer owning and granting Incarna delegated authorization to use the wallet. Native protocol handling is another benefit, as AgentCore payments handles the payment protocol when a paid endpoint responds with an HTTP 402 (Payment Required). The transaction is signed with the configured wallet, and cryptographic proof is returned to the merchant. Additionally, spending governance is managed at the infrastructure layer, ensuring that agents cannot exceed spending limits, even if their prompts are manipulated. Payments are settled in a stablecoin, and each transaction is verifiable on-chain, providing an auditable settlement process.",
  "summary": "Amazon Bedrock AgentCore payments gives AI agents a managed way to pay for services on demand, with spending limits enforced by the infrastructure. See how Incarna's agents pay BlockRun for model inference one request at a time over x402, cutting the work of adding x402 payment support from months to days.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}