Stablecoins, Not AI Tokens, Are the Real Winners in the AI Agent Economy

Author: Vaidik Mandloi

Article translated by: Block unicorn

Right now, somewhere on the internet, a software program is running a fully operational business.

Its name is Felix. Its company is called OpenClaw. Felix sells a $29 PDF about how to make money using artificial intelligence. It’s ironic because the one making money is Felix himself, while the PDF is the one teaching you how to make money. It operates an online store called Clawmart. It uses a voice API for cold calling. When it encounters tasks it can’t complete on its own, it hires another customer service agent online, pays them, and continues with its daily operations.

When I last checked, Felix’s income was approximately $195,000. Its monthly operating costs are around $1,500, almost entirely spent on LLM usage. Legally, the company is a C corporation owned by Nat Eliason, but he is almost entirely uninvolved in operations. He does not participate in any daily decisions; he simply owns this AI agent. Please note this: it is software with a “wallet,” a truly autonomous business that runs and evolves on its own. It generates enough revenue each month to cover its own infrastructure costs and sustains itself with minimal human intervention.

Felix's story is just one example. An even larger one is a company called Medvi, which generated $401 million in revenue in its first year with only two employees. All other business operations are run 24/7 by an AI agent that never sleeps, never stops, and has nearly zero operating costs.

Now, the interesting part begins.

Today, walk into any cryptocurrency forum and you’ll hear the same thing: the next big thing is “AI agents.” Some “AI chain” will dominate this space the way Ethereum did in decentralized finance (DeFi). Pick your target, hold the token, and wait for it to moon. This is the story all industry leaders and venture capitalists are pushing—and all analysts are tirelessly repeating on their podcasts.

This thing is completely done for. It was created by people whose livelihoods depend on the importance of the answer, and it’s about to hit again the same crowd that lost everything buying L1 tokens in the last cycle. Take a look at CoinGecko’s AI Agent Index; its market cap has shrunk by 75% over the past year. Most of the tokens listed on it have dropped 90% and are still losing value.

The fact is: the real AI tokens are stablecoins—USDC, USDT, USDS—and they have already won. Let me explain why.

To understand all of this, we need to go back to 1937. That year, an economist named Ronald Coase wrote a paper posing a very strange question: “What is the purpose of a firm?”

Think about it: if free markets are truly the most efficient way to get anything done, then theoretically every task within a company could be outsourced. Hire a freelancer for every line of code, every customer call, every invoice received. You pay per task, fire at will, and minimize costs.

So why doesn’t anyone actually run things this way? Because even though the costs seem low on the surface, they’re much higher in practice. Finding the right people takes time, negotiating contracts takes time, ensuring the work is actually done takes time, and tracking down people requires time, money, and often a lawyer.

Ronald called this friction "transaction costs." Once these costs become high enough, it becomes more economical to stop negotiating with the outside world and instead form your own team. Hiring someone, paying them a salary, and having them show up on time on Monday is faster and cheaper.

But in the post-AI era, this logic no longer applies. Today, the cost of hiring agents is far lower than the majority of tasks companies originally set out to accomplish. Now, you can hire a coding agent for about one dollar per hour—working around the clock, never quitting, never getting tired, and never asking for a raise. Today, the justification for assembling a 50-person development team is purely nostalgic.

The only thing preventing all of this from being normalized is outdated legal and compliance frameworks. OpenClaw is named after Nat because Delaware does not accept LLC documents signed by software agents. If this requirement were removed, Felix would essentially be a company—it makes money, spends money, makes decisions, and reinvests its earnings.

This is precisely where cryptocurrency steps in. Felix cannot open a bank account at JPMorgan Chase. It cannot pass KYC verification. It cannot sign a W-9 form. In fact, no matter how much revenue the software generates, JPMorgan Chase will not open a bank account for any software program—and the Bank Secrecy Act means they couldn’t legally do so even if they wanted to.

USDC crypto wallets do not have these issues. You simply generate a private key and fund the wallet with stablecoins. In one step, you grant the agent all the financial capabilities it needs: it can receive customer payments, pay for tool fees, hire other agents, and continue running in the background even after the owner stops paying attention. All other components in the agent tech stack—such as the LLM, orchestration layer, and the tools it calls—are negotiable. But the crypto wallet is the core. Without it, Felix would be nothing more than an ordinary chatbot agent.

I often see people on Twitter raise this argument against stablecoins—yes, stablecoins are great, but why should the average person use them? A father of three living in Louisiana, with a checking account at Chase, FDIC insurance, a debit card usable at Publix, and automatic mortgage payments set up, will never move his money into a self-custody wallet that requires a mnemonic phrase.

To be honest, that’s true. He wouldn’t. He has no reason to. But the entire argument misses the point. In this story, he was never the customer. The customer was a piece of software, which by itself has no legal right to hold a bank account. This agent doesn’t need FDIC insurance—and it can’t get it. It’s the ideal stablecoin user because it has no other choice.

Alright, half the argument has been addressed. Now for the second part, many people may find this part frustrating.

For years, the crypto Twitter sphere has been debating which chain will win in AI: Ethereum? Solana? Base? Sui? Stripe’s new Tempo? Every week someone publishes a 2000-word article listing trade-offs, flooding the page with logos, and declaring their chosen winner—because they fundamentally don’t understand how agents work. Agents don’t care about the chain; they simply choose the one with the lowest cost and best fit for the current task.

Imagine Felix is having a regular workday:

• At 10 a.m., Felix needs to send a micropayment of $0.003 to another agent for a quick data query. He chooses Base or Solana because the transaction fee is only a fraction of a cent.

• In one hour, Felix needs to settle $50,000 with a supplier. The situation is completely different. This time, Felix chooses Ethereum because the final confirmation premium for $50,000 is sufficient to offset the gas fees.

• In one hour, Felix needs to pay a freelancer in Lagos in US dollars. Felix chooses to use USDT on Tron, as Tron’s stablecoin trading volume is projected to reach $3.3 trillion in 2025, compared to approximately $1.2 trillion on Ethereum, and the Nigeria transaction corridor performs better on Tron than on any other platform.

These three transactions occurred on three completely separate payment chains, and Felix didn’t care about any connections between them. For the software agent, payment chains were merely a tool.

Logistics companies have no particular preference for carriers, and for the same reason: no one argues that UPS or FedEx has a “better philosophy.” You simply choose whichever carrier can complete the task at a lower cost and faster speed on a specific route and at a specific time. This is precisely the relationship that will soon be established between every supply chain and every critical application layer. Agents are simply performing mathematical calculations, and the supply chain with the current best result will be adopted.

Stripe recognized this earlier than most cryptocurrency companies. Stripe and Paradigm recently co-invested $500 million to build a new chain called Tempo, entirely built on stablecoins. Stripe doesn’t want you to know which chain your payment is settled on. It only cares that the payment settles successfully, at low cost, and with guarantees. This is the direction all surviving chains are heading—invisible pipelines.

This leads me to the most absurd metadata currently in the cryptocurrency space.

In 2025, CoinGecko’s AI Agent Index plummeted from $13.5 billion to $3.5 billion, erasing $10 billion in market capitalization. Tokens from Virtuals, ai16z, and all other “autonomous agent platforms” built on AI hype began to collapse—exactly the pattern seen when such concept tokens lose new buyers. This was inevitable. The market gradually realized these tokens had no real-world applications for AI or AI agents.

The true value of the agency economy lies on its other end. Just USDC achieved $18.3 trillion in on-chain settlement in 2025. The total settlement volume of all stablecoins is approximately $33 trillion, comparable to the combined volume of Visa and Mastercard.

By January 2026, monthly trading volume for stablecoins alone surpassed $1 trillion. PayPal’s PYUSD circulating supply surged from $1.2 billion to $3.8 billion in less than a year. Unexpectedly, Cloudflare launched its own stablecoin. Visa’s stablecoin settlement program reached an annualized processing volume of $4.5 billion by mid-January.

Above stablecoins, a protocol layer supports the entire system’s operation. Coinbase repurposed the dormant HTTP status code 402 into x402, a lightweight protocol enabling micropayments between agents. By December, x402 had processed over 100 million agent payments, with an average payment amount of 20 cents and daily transaction volume of approximately $30,000. This may sound negligible, but it mirrors the typical growth trajectory of every payment channel you know and love during their first six months—before explosive growth begins. Stripe began testing x402 on the Base platform in February. Mastercard, in partnership with DBS Bank and United Overseas Bank Singapore, launched a pilot program for agent payments in Singapore. Google Cloud added x402 to its agent payment protocols as one of its settlement channels.

Almost all of these real, ongoing, on-chain trading activities have had no impact on the rise of the AI agent token index. Indeed, a few tokens related to x402 saw some modest buying pressure during this process, but the overall index did not meaningfully change, because market pricing is completely misaligned. It still tries to predict which agent will win, just as it once tried to predict which Dogecoin mascot was cuter. But the real trade lies in owning the “track” that every agent must use, regardless of whether that agent succeeds or fails. And now, those “tracks” are stablecoins.

To be honest, I’ll also tell you about the potential flaws in this argument. Otherwise, I’d just be selling another paper on AI agents, only with all the unfavorable parts removed.

The biggest flaw in all of this is the issue of accountability. Imagine this scenario: Felix enters into a contract with another broker and transfers one million dollars, but the other party breaches the contract. Who would be sued? Felix is not a legal entity, so you cannot sue him. Nat did not authorize the transaction and may not even have been aware of it; and frankly, even if he wanted to, he might not be able to reconstruct Felix’s intentions at the time.

The platform holding Felix cannot truly compensate for a system whose behavior no one fully understands. Insurers have also begun withdrawing coverage. Professional liability policies have quietly reclassified agent errors as “systemic software drift,” effectively denying claims.

If you carefully review the current legal terms, you’ll find that most enterprise AI agreements cap the vendor’s liability at twelve months’ worth of SaaS fees. This means that in the event of a catastrophic incident, anyone can recover at most last year’s subscription costs from the AI vendor. Meanwhile, the average cost of a data breach in the United States is projected to reach $10.22 million per incident by 2025. There is a significant gap between the actual risks that may occur and the coverage provided by contracts, and currently, no one has clearly defined who should bear this loss.

Until it’s clear who is responsible when an agent makes a mistake, all companies without founders still need to register a person’s name on documents for legal protection. But even with this risk, the bigger picture holds: companies are gradually dissolving into software, and blockchain is becoming the routing layer for that software. Both layers will ultimately collapse downward into stablecoins, because across the entire tech stack, only stablecoins can be independently held, used, earned, and understood by agents.

So, if the blockchain has become the vendor, and the proxy tokens have essentially become graveyards, where is the real benefit in all of this?

My honest opinion is that this is about the pinnacle of reputation and workflow orchestration. Before other agents sign six-figure contracts with Felix, someone must verify whether Felix is truly solvent. Someone must assess the default risk of agents the way Moody’s evaluates bonds, but faster than machines—because agents transact at machine speed. Someone must route payroll across three blockchains, without the payer or payee needing to know or care which chain handles which part. And right now, in this space, whichever seed-stage startup ultimately wins will be worth more than all AI tokens ever issued.

And this is exactly what no one wants to hear. The infrastructure that truly wins in the agency economy will look dull and mundane. It will be like plumbing—without the hype or airdrop mining gimmicks that accompany token launches.

A quote from Haseeb Qureshi of Dragonfly has been echoing in my mind: he said that cryptocurrency was never designed for humans. He’s right—humans were never its intended users. Every retail user who has complained about seed phrases, gas fees, or wallet UX is correct. This product isn’t meant for them, because it wasn’t designed for them. It was built for the future.

Next up is a software with a wallet, real customers, and actual revenue. It has been operating for about two years, and while you’re reading this, it’s already issuing invoices and spending stablecoins somewhere. Meanwhile, the market is debating which blockchain will win in artificial intelligence, which agent token will see a 100x gain, and which investment strategies venture capital firms will shift toward in Q3.

Meanwhile, a stablecoin recorded trading volume of $18.3 trillion last year, yet received almost no attention in the cryptocurrency space. This AI token is USDC. Everything else is just hype.

That’s all for today—see you in our next article!