On August 18, AWS made Amazon Bedrock AgentCore Payments generally available — which means enterprise AI agents can now carry a wallet, pay for services on their own, and keep working without stopping to ask for approval.

The announcement, built in partnership with Coinbase and Stripe, marks a meaningful step toward AI agents that behave like autonomous workers rather than sophisticated chatbots. An agent that can pay for what it needs is an agent that can actually get things done end-to-end.

What AgentCore Payments Actually Does

The mechanics work like this: when an AI agent running on Amazon Bedrock encounters a paid resource — a premium API, an MCP server that charges per call, a paywalled data feed — it receives an HTTP 402 response. Normally, that’s where automation dies.

With AgentCore Payments, the agent doesn’t stop. AgentCore handles the full payment cycle in the background: it negotiates with the resource, authenticates the wallet, executes the payment, and sends confirmation back to the endpoint. The agent continues its task without interruption.

Developers connect either a Coinbase CDP wallet or a Stripe Privy wallet to their agent configuration. They set session-level spending limits — how much the agent is allowed to spend per task run. The infrastructure enforces those limits. An agent cannot exceed them.

Two Payment Protocols Now Supported

When AgentCore Payments launched in preview in May, it only supported the x402 protocol, an HTTP-native standard for machine-to-machine payments backed by Coinbase and Visa. That limited who could actually accept agent payments.

At GA, AWS added support for the Machine Payment Protocol (MPP), co-authored by Stripe and Tempo. MPP gives agents access to a wider ecosystem of merchant endpoints. The practical result: more of the tools and services your agents need to interact with can now be paid for automatically, without custom integration work.

Both protocols use stablecoin (USDC) for transactions, which keeps settlement fast and removes currency conversion overhead from the equation.

Observability Included

One concern with autonomous spending is visibility. If an agent is making payments on behalf of your business, you need to know what it spent, on what, and when.

AgentCore Payments integrates with AgentCore Observability, giving you an audit trail of every transaction. You can see which agent spent what, on which resource, during which task. This is the kind of governance infrastructure that makes executive sign-off on autonomous agents a real conversation rather than a non-starter.

The service is available in four AWS Regions: US East (N. Virginia), US West (Oregon), Europe (Frankfurt), and Asia Pacific (Sydney).

What This Means for Business

The agent payments story has been building across the industry for most of 2026. Visa launched Agentic Ready infrastructure, Nevermined demonstrated x402-powered card payments for agents in April, and the Rain Agentic Payments Alliance has been pushing for broader merchant adoption.

AWS putting this into production-grade infrastructure changes the calculus for enterprise deployments.

Here’s the practical shift: until now, building AI agents that could pay for external services required custom payment plumbing — connecting wallets, handling 402 errors, managing auth flows, building audit logs from scratch. Most enterprise teams skipped it. Agents were limited to free-tier APIs or resources that were already licensed upfront.

AgentCore Payments removes that friction. You configure a wallet, set spending limits, and your agent gains access to the entire ecosystem of paid tools and services it might need.

For businesses running agent-based operations — automating procurement research, competitive intelligence, data enrichment, or anything that requires pulling from paid data sources — this unlocks a category of automation that wasn’t practical before.

The governance angle matters too. Session-level spending limits mean your finance team can set hard caps on what any given agent workflow is allowed to spend. The agent operates within a pre-approved budget, automatically. That’s a meaningful difference from giving someone a corporate card and hoping they expense things correctly.

The Broader Shift

Enterprise AI agent adoption has been accelerating faster than most expected. Salesforce’s Agentic Enterprise Index found organizations went from running an average of five agents to thirteen between early 2025 and April 2026. The skill set of an average agent tripled in the same period.

As agents take on more complex, multi-step workflows, their ability to access and pay for the tools they need becomes a real operational constraint. AgentCore Payments is AWS’s answer to that constraint — and it’s now in production.

---

If you’re thinking about how to deploy AI agents across your business operations, the question is no longer whether agents can do the work. It’s whether your infrastructure lets them operate without constant human handoffs. Autonomous payments is one more step toward agents that genuinely run end-to-end.

Talk to the Enterprise DNA team about building an AI agent workforce for your business.