“When agents make the majority of financial decisions, economic decisions, how do they transact with each other?” That question, posed by Chappy Asel at Consensus Miami this week, frames what might be crypto’s most plausible path to mass adoption: stop trying to onboard humans.
Asel, founder of The AI Collective (a nonprofit AI community with over 200,000 members across more than 150 chapters), spent years at Apple working on Vision Pro and early Apple Intelligence initiatives before pivoting to AI community building. His thesis is straightforward: autonomous software agents will need payment infrastructure that humans never really demanded, and crypto happens to be the only system designed for exactly that use case.
Machines Don’t Need Tutorials
The crypto industry has spent a decade trying to simplify wallet interfaces, explain seed phrases, and make MetaMask less terrifying to normies. The results speak for themselves: Bitcoin remains a speculative asset for most retail holders, and DeFi usage has never escaped a core user base that already thinks in hexadecimal.
AI agents sidestep every one of these friction points. Software doesn’t get confused by gas fees. It doesn’t lose recovery phrases. It doesn’t need a YouTube tutorial to understand that a smart contract is just code that executes when conditions are met. If machines become meaningful economic actors (and that’s still an if), they represent a user class that crypto’s existing infrastructure already serves well.
“You want them to be highly systematic, mechanistic,” Asel said during his panel. “You want very small, micro transactions. You want very low latency.”
That description maps precisely to what stablecoins and programmable blockchains offer. Settlement happens around the clock. Smart contracts execute without human approval loops. Transaction sizes can be fractions of a cent. For an AI agent managing a fleet of autonomous delivery drones, each making hundreds of routing and payment decisions per hour, batching those into daily ACH settlements through traditional banking would be architecturally absurd.
The Gap Between Narrative and Revenue
Here’s where the pitch hits friction: almost none of this is happening yet.
Asel acknowledged the gap directly. AI agents remain nascent technology, and most companies deploying them still route payments through centralized APIs and conventional systems. Attempts to build dedicated agentic payments infrastructure have produced conferences, whitepapers, and Twitter threads, but little meaningful commercial activity.
The term “agentic payments” itself has achieved a kind of memetic success that outpaces actual usage. Asel noted that even his friends who know nothing about blockchain have heard the phrase. That’s a warning sign as much as a bullish indicator. Crypto has a long history of narrative cycles that front-run adoption by years (or forever). “Remittances” was the killer app circa 2014. “Banking the unbanked” peaked around 2017. Neither delivered at scale.
The difference this time, proponents argue, is that the underlying technology finally exists. Large language models can now orchestrate multi-step workflows, access APIs, and make decisions within defined parameters. Whether they need crypto-native payment rails or will simply use Stripe remains an open question the market hasn’t answered.
For founders trying to build in this space, Asel’s advice was less strategic than temperamental: experiment. “When the world is more uncertain than it ever has been… things will only get crazier,” he said. “That warrants that you are spending more and more time playing around with the new technology.”
That’s reasonable guidance for anyone with runway to burn, though it doesn’t tell you much about where actual demand will materialize.
Compute, Not Models, Is the Constraint
If agentic payments remain speculative, Asel pointed to a more immediate overlap between crypto and AI: physical infrastructure.
“A lot of people will tell you, oh, it’s the models aren’t good enough,” he said. “It’s none of that. It’s literally compute, data centers, energy that is driving pretty much all decision-making in AI right now.”
This framing reflects a broader shift in how the AI industry thinks about competitive advantage. Access to Nvidia chips, power purchase agreements, and data center capacity now matter more than incremental model improvements. OpenAI, Anthropic, and their competitors are increasingly defined by their ability to secure infrastructure, not just train better weights.
Crypto firms have noticed. Several Bitcoin miners have spent the past year repositioning toward AI hosting and high-performance computing. The logic is straightforward: mining operations already have the three things AI workloads need most. They have relationships with power providers (often with favorable rates negotiated when BTC prices were higher). They have cooling infrastructure designed for heat-dense compute. And they have real estate in locations where data centers can actually get built and permitted.
Whether mining infrastructure can actually serve AI workloads at competitive economics remains unproven. Mining rigs and GPU clusters have different power profiles, cooling requirements, and network needs. But the bet is that the hard parts (land, power, permits) transfer, even if the easy parts (racking servers) require rebuild.
For those tracking crypto’s infrastructure plays, our derivatives dashboard shows how mining company valuations have diverged based on their AI pivot narratives, with funding rates and open interest reflecting market skepticism about pure-play mining economics at current BTC prices.
A User Base That Thinks in Code
The deeper argument Asel made wasn’t about payments or infrastructure specifically. It was about fit.
Crypto has always been a system designed by engineers for other engineers. Its abstractions (public-key cryptography, deterministic state machines, merkle trees) are native to people who already think computationally. The entire history of “improving UX” has been an attempt to paper over that reality for humans who don’t.
AI agents don’t need that paper. They don’t need MetaMask redesigns or friendly onboarding flows. They can interact with smart contracts directly, manage private keys programmatically, and parse transaction data without UI translation layers. In a sense, autonomous software is the user base crypto was accidentally built for.
That doesn’t mean adoption is inevitable. Plenty of enterprise software serves machines rather than humans, and most of it runs just fine on traditional databases and payment APIs. The question is whether crypto’s specific properties (censorship resistance, programmability, global settlement, permissionless access) matter for agentic use cases in ways they haven’t mattered for human use cases.
If an AI agent managing a supply chain needs to pay a warehouse in one jurisdiction and a shipper in another without waiting for banking hours or approval workflows, stablecoin rails offer genuine advantages. If the same agent just needs to move dollars between two corporate accounts at scheduled intervals, Stripe probably works fine.
The bet Asel and others are making is that autonomous software will face enough edge cases (cross-border, 24/7, micro-sized, programmatic) that crypto rails become default rather than exceptional. But edge cases have a way of staying at the edges, and most economic activity remains stubbornly normal.
What Actually Has to Happen
For agentic payments to matter, several things need to occur in sequence. AI agents need to become capable enough that businesses trust them with financial decisions. Those agents need to encounter payment frictions that traditional rails can’t solve. And crypto infrastructure needs to be mature enough (fast enough, cheap enough, legally clear enough) that it becomes the path of least resistance.
Each of those steps has skeptics. Model capabilities are improving rapidly, but reliability for high-stakes autonomous decisions remains questionable. Payment frictions exist, but most businesses design around them rather than seeking new rails. And crypto infrastructure, while improved, still carries regulatory uncertainty in most major markets.
Asel’s view is that experimentation now positions founders for whatever the market eventually demands. That’s a defensible stance, even if it’s not a prediction. The honest version of the thesis is: we don’t know if machine-to-machine commerce will be big, we don’t know if it will need crypto, but if both things happen, being early matters.
Crypto’s consumer problem has always been that normal people don’t want to think about keys, chains, or consensus mechanisms. AI agents might be the constituency that actually prefers it that way.
Related Reading
- Seed phrase security, start to finish
- Technology news
- More on AI Agents
- More on Stablecoins
- More on Agentic Payments
Source Material
This content is educational, not financial advice. Digital asset investments can lose value. Research thoroughly before investing.
