The third use case.

I've long argued that the crypto industry has only found two real, value-add use cases in its nearly two decades of existence. The first was Bitcoin's initial product-market fit as a censorship resistant store of value. The last few years has seen the emergence of a second: the integration of stablecoin technologies in the larger financial ecosystem as an alternative transfer and settlement layer for dollar-denominated value. Despite some initial reservations, I'm now convinced that agentic payments and commerce will be the third.

In defense of my skepticism, the most common scenarios given for agentic payments tend to be the least interesting (and least likely) ones. We're told that humans will no longer need to book flights themselves, shop on Amazon, or remember to buy anniversary flowers. Instead of paying for access to a paywalled Times article or Messari's market insights, they'll delegate these responsibilities to an agent. Certainly these are things that will happen. But I don't see them as the primary drivers of growth here.

Transient and consumptive

The real use case will be paid access to assets or information that is both transient and consumptive. Think model inference, lambda invocations, and emerging categories like AI-first service providers; Exa AI's web search for agents² being a perfect example. There's a strong fit for crypto there, as the cost of invoking one of these services can be fractions of a penny. VISA, Mastercard, and the like cannot process transactions at such a low fee rate. An L2 like Base, on the other hand, charges $0.002 per transaction¹ - that's 50-100x cheaper than VISA - and offers real-time settlement and transaction finality. It "just works" with no need for KYC, minimum transaction sizes, or even persistent user identities.

That said, it's worth noting that it is also a boon to attackers. Irreversible settlement and no persistent identity means no chargebacks and no path for recourse. The properties that make these rails perfect for autonomous agents are the same ones that make agent compromise disastrous.

And agents do get compromised

Last month an attacker drained roughly $200k from an AI agent's wallet on Base with a simple tweet³. It contained a transfer instruction encoded in Morse code, which the agent helpfully decoded and executed. It was the first documented prompt injection against a live wallet, but it will certainly not be the last.

Who would really trust such non-deterministic, easily coerced systems to have direct access to their payment information and private keys? A lot of people apparently. Agentic Market tracks about 120k transactions per day, and a payment volume of $56,752.19 for June 1, 2026⁴. The industry is racing to close the gap between adoption and security: Google's AP2 protocol enforces signed payment mandates⁵ and Stripe subsidiary Privy is building policy-gated wallets with payment caps and allowlists⁶.

Real, bullshit, or ten years early

Working in this industry, it's often hard to tell what's real, what's bullshit, and what's ten years too early. That said, I'm certain that the category of agentic payments, while still nascent, will be the next big driver of long-term crypto adoption.

For now, if anyone has any real world experiences where they've seen value in agentic payments, either as a vendor or consumer, shoot us an email. We would love to discuss.