Agentic Finance (AiFi) on AWS, Powered by Coinbase: GA today with AgentCore Payments

By: Matt Wong and Josh Nickerson

TL;DR: Any agent hosted on Amazon Bedrock AgentCore can now discover and pay for services autonomously, with no custom payments infrastructure required. Companies, like Travala, are already using these payments, powered by Coinbase Developer Platform and x402, to give users and employees new experiences using Agentic Finance (AiFI). 

Agentic finance needs three things to work at scale: a place for agents to live, a payment rail they can trust, and an open way to find what to pay for. Amazon Bedrock AgentCore is AWS's flagship managed runtime for production AI agents, making it the default choice for enterprises building on AWS. With AgentCore Payments, that runtime now has a built-in way to transact.

Developers can now provision CDP wallets for their agents and users, and users can give agents permission to spend on their behalf within guardrails they control. When the agent needs a paid resource, it discovers and pays for it through x402, the open protocol powering a curated  list of services in the x402 Bazaar.

AWS brings the scale: any agent hosted in AgentCore gets this by default. Coinbase brings the trust: battle-tested CDP wallets. And x402 brings the openness: autonomous discovery across a growing, permissionless catalog of services. An agent can go from "I need this" to "I paid for it, and I'm still working," with no developer wiring signing, wallet infrastructure, or spend controls by hand.

One Integration, Two Very Different Use Cases

This integration gives any company a strong foundation of features to start up quickly, securely and easily,

• A wallet without the setup. AgentCore Payments makes it easy to provision CDP wallets for agents and users, so there are no private keys or seed phrases to manage.

• Native x402 signing. When an agent needs to pay, the payment is signed and sent without custom protocol plumbing.

• User-set guardrails. Users grant agents permission to spend from their wallets, with configurable limits on how much the agent can spend.

• Isolated credentials. Wallet credentials stay isolated from the agent's runtime memory, keeping keys out of reach of the model.

• Fast, low-cost settlement. Payments settle onchain in stablecoins.

The result: an agent knows exactly how much it's allowed to spend before it ever acts, then transacts autonomously within that limit.

Use Case 1: Travala's Travel MCP Books Trips Inside Claude

Travala's Travel MCP, introduced in June, is an agentic travel protocol that lets AI agents search, book, and pay for trips autonomously onchain. Travelers can book and manage trips conversationally in Claude, and developers can integrate the Travala Travel MCP into their own agents.

AgentCore payments is the infrastructure underneath the booking flow, connecting to a CDP wallet, enforcing payment limits, and handling security in the background so the traveler never sees the complexity. Booking one of Travala's 2.2 million-plus properties looks like this:

1. Connect to the Travala Travel MCP in Claude.

2. Search for a stay.

3. Grant the agent spending permission through CDP, which provisions a wallet automatically.

4. Fund the wallet with USDC via Coinbase and complete the booking in Claude.

Developers who integrate the Travala Travel MCP into their agents also automatically earn a 10% cbBTC rebate on every successful booking as part of Travala’s developer rebate program. 

"Integrating AgentCore payments into the Travala Travel MCP streamlines the agentic booking process for our 2.2 million properties globally. This seamless integration offers a secure way to handle agentic transactions with built-in payment limits, improving the overall agentic booking experience for our customers." – Juan Otero, CEO of Travala

Use Case 2: The Agentic Economy, One Slack Message Away

Within Coinbase, we wanted any employee to ask for what they need right in Slack, a tool they already use every day. So we built an agent that can find and pay for the right service automatically, without them ever needing to know x402 exists. 

This agent puts the full x402 Bazaar behind every answer: ask something, and the agent searches thousands of x402-enabled services, decides which ones can actually help, and pays for them via a prefunded CDP wallet. It works across four layers:

• Interface. Employees ask questions in Slack.

• Reasoning. An LLM evaluates the question, searches the x402 Bazaar for services that could help answer it, and decides whether a paid service is worth using or whether existing knowledge is enough.

• Payment. When a paid service is the right call, a tool call triggers AgentCore Payments, which signs the x402 payment through the connected wallet against a pre-set spending limit.

• Settlement. The resource server (x402 seller) verifies and settles the payment, and the result flows back into the same reply. No separate approval step, and no context switching.

For the pilot, we chose one shared, pre-funded wallet rather than one per person. That removed the sign-up and funding step entirely, speeding up adoption while keeping spend governed on every request. Per-user wallets remain an option as usage grows.

What This Unlocks

These two examples are the same pattern wearing different clothes. One is a consumer app that books hotels, the other is an internal tool that finds and pays for whatever an employee needs. Both let an agent pay on its own, and both stay inside a spend limit set by the wallet before the agent ever acts. Once agentic finance is a managed building block instead of a custom project, the question stops being "how do we let an agent pay" and becomes "what should we let it pay for."

Get Started

• Join the x402 Foundation Slack to see a live demo: slack.x402.org

• Read the docs: https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/payments.html

• Learn more about AgentCore payments: https://aws.amazon.com/blogs/machine-learning/amazon-bedrock-agentcore-payments-is-now-generally-available-enabling-agents-to-transact-safely-and-autonomously-at-scale/