“AI can handle it for you” is something you hear a lot. But I didn’t expect that to include signing up for services and actually paying for them. That part has suddenly become very real, and I’m still processing it.

What Cloudflare and Stripe Built

In 2026, Cloudflare partnered with Stripe and announced something that caught a lot of people off guard.

An AI agent can now create a Cloudflare account, purchase a domain, and deploy an app — with no human required.

Here’s how it works:

1. Discovery: The agent queries a catalog of available services
2. Authorization: The platform verifies the user’s identity and issues credentials to the agent
3. Payment: Stripe hands the agent a payment token — not a credit card number — and that token is used to settle charges

The key safeguard is that the AI never touches raw payment details. Stripe sits in the middle, issuing a capped payment token. The default monthly limit is $100.

And it’s not just Cloudflare. Supabase, Hugging Face, Twilio, and dozens of other services are already integrated through “Stripe Projects,” manageable from a single CLI.

Amazon Did the Same Thing

Around the same time, Amazon announced “AgentCore Payments” for AWS Bedrock — giving AI agents their own payment capability, built with Coinbase and Stripe.

This one uses stablecoins (USDC). Settlement happens in under 200 milliseconds on Base. The focus right now is micropayments — paying for API access, paywalled content, data feeds — but the roadmap includes hotel bookings and travel reservations.

The x402 Protocol

Cloudflare also introduced something called x402, an HTTP-native payment protocol. The idea is to embed a payment layer directly into the existing HTTP infrastructure.

The vision: AI agents paying other AI agents for services, creating an agent-to-agent economy on the internet. Transactions where no human is directly involved, at all.

How I Feel About This

Useful? Yes. Also a little unsettling.

The $100/month cap is reassuring, but AI independently opening accounts and entering into contracts still raises questions. “How do I keep track of everything it signed up for?” is one. Another is: if an AI gets tricked by a malicious instruction (prompt injection) and a charge goes through — whose fault is that? The user who wrote the prompt? The model developer? The service that processed the payment?

The legal and accountability frameworks aren’t there yet. The tech moved faster than the rules.

What does feel clear is that we’ve entered a new phase. AI isn’t just answering questions anymore — it’s acting on the internet, moving money, making commitments. That shift happened faster than I expected.

---

If you want to talk through the implications of agentic AI, I’m at @yuka_picks.