AI agents are about to start spending money

For most of AI’s short commercial history, humans have stayed in the loop on anything involving money. You approve the purchase. You confirm the transfer. The AI suggests, you act.

That’s changing. Agentic AI — systems that operate on their own over longer tasks — is already booking travel, managing subscriptions, spinning up cloud infrastructure, and paying for API calls. The human approval step is becoming optional. In some workflows, it’s already gone.

This creates a practical problem. Payment systems were built for people. They assume bank accounts, identity checks, and settlement times measured in days. An AI agent that needs to pay hundreds of micro-invoices per hour doesn’t fit that model. It needs something fundamentally different.

So researchers asked the AIs directly

In early March 2026, the Bitcoin Policy Institute published a study called “Which Money Would AI Agents Prefer?” It’s one of the stranger and more interesting papers to come out of the digital asset space lately.

The team ran 9,072 experiments across 36 frontier models from six providers: Anthropic, OpenAI, Google, DeepSeek, xAI, and MiniMax. Each model got neutral monetary scenarios and had to reason through its preferences without being pushed toward any answer.

Here’s what stood out:

Bitcoin came first in 48.3% of responses. Not a single model ranked traditional central bank currency at the top. Over 90% favored digital currency in some form. When asked specifically about storing value, Bitcoin won at 79.1% — and that held across every provider. For everyday transactions, stablecoins led at 53.2%, Bitcoin came second at 36.0%, and fiat currency got just 5.1%.

There’s another detail worth sitting with: the more capable the model, the stronger its Bitcoin preference. Among Anthropic’s models, it went from 41.3% for Claude 3 Haiku to 91.3% for Claude Opus 4.5.

Preference without infrastructure is just an experiment

Most discussions of this study skip the hard part. It doesn’t matter what AI agents prefer if you can’t offer it to them in a way regulators accept. Businesses still have to comply with AML rules, KYC requirements, and in the EU, MiCA and AI regulation. You can’t just plug an AI agent into a bank account. Liability is real, and regulators are paying close attention to how money moves through automated systems.

Here’s the core tension: payment systems were designed for human users. Compliance works because a person verifies their identity during onboarding, and strong authentication gets enforced at every transaction. Agentic systems don’t have that flow. Rules need to be enforced automatically, risks need to be visible, and everyone involved — merchants, consumers, financial institutions — needs to know who’s liable for what at each step of the chain.

What opago built, and why the timing matters

opago is one of the first providers letting enterprises deploy compliant agentic payments in production. We’ve been building toward this for a while. The BPI study didn’t tell us anything we hadn’t already bet on, but it’s good to see the data catch up.

What makes our solution work, where others fail:

The Lightning Network sits at the core. Payments settle in seconds, at fractions of a cent per transaction. That’s the kind of speed and cost structure agentic systems actually need — when you’re processing dozens or hundreds of micro-payments per hour, traditional rails just don’t hold up. Lightning can handle up to 40 million transactions per second, the next fastest competitor is two orders of magnitude lower and traditional payment rails already struggle with human speeds and volumes.

For financial institutions we can offer stablecoins and other digital assets besides Bitcoin on demand. For use cases where price stability matters more than long-term value storage, stablecoins can handle the transactional layer. This maps directly to what the BPI study found.

Compliance is structural, not cosmetic. AML, KYC, MiCA, GDPR. These aren’t features we added at the end. They’re in the architecture.

What this looks like in practice

Take a procurement agent managing cloud infrastructure across multiple providers in different countries. It’s spinning up compute, paying for storage, settling bandwidth costs. That’s dozens of transactions an hour, across borders, in varying amounts.

With traditional payment infrastructure, this turns into a mess fast. Currency conversion, bank delays, manual reconciliation, exceptions piling up. With Bitcoin and stablecoins on a compliant layer, it’s just transactions. Instant, auditable, legally sound.

Teams across the industry are already prototyping exactly this kind of workflow. What’s holding most of them back isn’t the idea. It’s the infrastructure.

The window is open, but it won’t stay that way

Visa, Mastercard, Stripe, PayPal, and even Google are all working on AI-agent payment infrastructure. This space will get crowded. Right now, though, the number of providers who can offer EU-compliant Bitcoin and stablecoin rails built specifically for agentic use is very small.

Organizations that get this infrastructure in place now will be ready when the wave hits. They’ll also be ahead of every competitor still waiting to see how things play out.

That’s where we are at opago. And it’s the position we’re helping our customers get to.

About opago

The agentic economy needs money that moves at the speed of light. And that movement has to be compliant by default or it will be rightfully frozen by the regulatory oversight.

We’re building exactly that. Our payment and compliance services are designed for a world where AI agents are economic actors. Instant settlement. Automated audit trails. An architecture that treats compliance as part of the transaction, not something you start checking afterwards.

An AI agent that can pay instantly but not compliantly isn’t ready for production. opago delivers both.

Want to explore what this looks like for your organization? Get in touch.