USDC is the only AI token

Author: Vaidik Mandloi

Translation by: Block unicorn

At this moment, somewhere on the internet, a piece of software is operating a complete business.

Its name is Felix. Its company is called OpenClaw. Felix sells a PDF for $29, which is about how to make money using artificial intelligence. This is ironic because the one making money is Felix itself, while the PDF is what teaches you how to make money. It operates an online store called Clawmart. It conducts telemarketing through a voice API. When it encounters work it cannot complete, it hires another customer service representative online, pays them, and continues with its daily tasks.

The last time I checked, Felix's revenue was about $195,000. Its monthly operating cost is around $1,500, almost entirely for the use of LLMs. Legally, this company is a C corporation, and the owner is Nat Eliason, but he is hardly involved in operations. He does not participate in any daily decisions; he simply owns this AI agent. Note this. This is software with a "wallet," a truly autonomous and continuously evolving business. It can pay for its own infrastructure costs every month. It can sustain itself with almost no human intervention.

Felix's story is just one example. There is a larger case, a company called Medvi, which achieved $401 million in revenue in its first year of operation with only two employees. The rest of the company's operations are run by an AI agent that works around the clock, tirelessly, and with almost zero operating costs.

Now, here comes the interesting part.

Nowadays, if you walk into a cryptocurrency forum, you will hear the same thing: the next hot topic is "AI agents." Some "AI chain" will lead the way in this field just like Ethereum did in defi-119">decentralized finance (DeFi). Pick your target, hold the tokens, and wait for them to skyrocket. This is the story all industry leaders and venture capitalists are selling, and it is the refrain all analysts tirelessly repeat on podcasts.

This thing is completely doomed. Because it was invented by those whose work relies on the importance of answers, and it is about to hit hard again the group that lost everything in the last round of buying L1 tokens. Look at the AI agent index on CoinGecko; it has shrunk by 75% in market value over the past year. Most of the tokens listed above have dropped by 90% and continue to lose value.

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

Software is now a company

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 silly question—"What is the purpose of a company?"

Think about it: if the free market is indeed the most efficient way to get anything done, then theoretically every task within a company could be outsourced. Every line of code could be done by a freelancer, every customer call could be handled by a freelancer, every invoice received could be managed by a freelancer. You could pay by task, fire at any time, and minimize costs.

So why doesn’t anyone actually run things this way? Because even if the surface costs seem low, the actual operation is more expensive. Finding the right person takes time, negotiating contracts takes time, ensuring the work is actually done takes time, and tracking personnel takes time, money, and often lawyers.

Ronald called this friction "transaction costs." Once these costs are high enough, instead of bargaining with the outside world, it becomes more cost-effective to stop negotiating with the outside and build 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. Nowadays, the cost of hiring agents is far lower than the majority of tasks that companies originally undertook. Now, you can hire a coding agent for about a dollar/hour, working around the clock, never quitting, never tiring, and never asking for a raise. The rationale for forming a 50-person development team is purely nostalgic.

The only thing hindering the normalization of all this today 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 lifted, Felix would effectively already be a company. It makes money, spends money, makes decisions, and reinvests the money it earns.

And this is where cryptocurrency starts to take on responsibilities. Because Felix cannot open a bank account at JPMorgan. It cannot pass KYC verification. It also cannot sign a W-9 form. In fact, no matter how much revenue software generates, JPMorgan will not open a bank account for any software program, and the Bank Secrecy Act means that even if they wanted to, they could not do so legally.

USDC crypto wallets do not have these issues. You simply generate a private key and then fund the wallet with stablecoins. With just one step, you grant the agent company all the financial capabilities it needs. It can receive customer payments, pay tool fees, hire other agents, and continue to operate in the background once the owner is no longer paying attention. All other components in the agent tech stack, such as LLMs, orchestration layers, and the tools it calls, are negotiable. But the crypto wallet is the core. Without it, Felix would just be an ordinary chatbot agent.

I often see some opponents of stablecoins on Twitter making the argument—sure, stablecoins are nice, but why would ordinary people use them? A father living in Louisiana with three kids, a checking account at JPMorgan, FDIC insurance, a debit card that works at Publix, and an automatic mortgage payment setup would never move his money to a self-custody wallet that requires a mnemonic phrase to use.

To be honest, that’s true. He wouldn’t. He has no reason to do so. But the entire argument misses the point. In this story, he was never the customer. The customer is a piece of software that has no legal right to hold a bank account. This agent does not need FDIC protection. It also cannot obtain FDIC protection. It is the ideal user of stablecoins because it has no other choice.

Chains are now suppliers

Alright, half of the argument has been addressed. Now we move to the second part, which might anger many people.

The cryptocurrency Twitter community has been debating for years which chain will win in the AI space: Ethereum? Solana? Base? Sui? Stripe's new Tempo? Every week someone publishes a 2,000-word article listing various pros and cons, inundated with logos, ultimately naming their perceived winner. Because they fundamentally do not understand how agents work. Agents do not care which chain they are on; they will choose the chain that is the cheapest and most suitable for the current task.

Imagine Felix on a regular workday:

• At 10 AM, Felix needs to send a micropayment of $0.003 to another agent for a quick data query. Felix chooses Base or Solana. Why? Because the fees are just a fraction of a cent.

• An hour later, 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 enough to offset the gas fees.

• An hour later, Felix needs to pay a freelancer in Lagos in dollars. Felix chooses to use USDT on Tron because Tron is projected to reach $33 trillion in stablecoin transaction volume by 2025, while Ethereum is around $12 trillion, and Nigeria's transaction corridor performs better on Tron than on any other platform.

These three payments occur on three completely different payment chains, and Felix does not care about the connections between them. For the software agent, the payment chain is just a tool.

Logistics companies have no special preference for carriers for the same reason. No one argues about which is "better" between UPS and FedEx. You simply choose which can complete the task at a lower cost and faster speed on a specific route at a specific time. This is the relationship that is about to be established between every supply chain and every important application layer. Agents are just doing math, and the supply chain with the best current calculation results will be adopted.

Stripe realized this earlier than most cryptocurrency companies. Stripe and Paradigm recently co-invested $500 million to create a new chain called Tempo, built entirely on stablecoins. Stripe does not want you to know which chain your payment is settled on. It only cares that the payment is successfully settled, at a low cost, and with guarantees. This is the direction all surviving chains will take in the future—an invisible pipeline.

This leads to what I believe is the most absurd pricing of metadata in the current cryptocurrency space.

The AI token graveyard

By 2025, CoinGecko's AI agent index plummeted from $13.5 billion to $3.5 billion, evaporating $10 billion in market value. Virtuals, ai16z, and all the "autonomous agent platform" tokens that were hyped up by AI concepts began to collapse, which is the typical pattern for such conceptual tokens once they lose new buyers. This was bound to happen. The market is gradually realizing that these tokens have no actual application scenarios for AI or AI agents.

The true economic value of agents is reflected on the other end. Just USDC alone achieved $18.3 trillion in on-chain settlements by 2025. The total settlement amount for all stablecoins is about $33 trillion, comparable to the total of Visa and Mastercard.

By January 2026, the monthly transaction volume of stablecoins alone surpassed $10 trillion. The circulation of PayPal's PYUSD skyrocketed from $1.2 billion to $3.8 billion in less than a year. Unexpectedly, Cloudflare launched its own stablecoin. Visa's stablecoin settlement project reached an annualized processing volume of $4.5 billion by mid-January.

Above stablecoins, there are protocol layers supporting the operation of the entire system. Coinbase transformed an idle HTTP status code called 402 into x402, a small protocol allowing payments between agents. By December, x402 had processed over 100 million agent payments, with an average payment amount of $0.20 and a daily transaction volume of about $30,000. This sounds pitifully small, but it is precisely the typical growth trajectory of all the payment channels you know and love in the first six months, just before explosive growth begins. Stripe began testing x402 on the Base platform in February. Mastercard partnered with DBS Bank and UOB in Singapore to launch an agent payment pilot project. Google Cloud added x402 to its agent payment protocol as one of the settlement channels.

Almost all of this real, ongoing, on-mainnet transaction activity has had no impact on the rise of the AI agent token index. Admittedly, a few tokens related to x402 gained some small buying interest in the process, but the overall index did not truly change as a result. Because the market pricing is completely wrong. It is still trying to predict which agent will win, just like it once predicted which Dogecoin mascot was cuter. But the real trade is in owning the "tracks" that every agent must use, whether that agent is alive or dead. And right now, those "tracks" are stablecoins.

Cracks in the paper

To be honest, I would also tell you the potential flaws in this argument. Otherwise, I would just be selling another paper on AI agents, having removed all the unfavorable parts.

The biggest loophole in all of this is liability attribution. Imagine this scenario: Felix contracts with another broker and transfers a million dollars, but the other party defaults. So who gets sued? Felix is not a legal entity, so you cannot sue him. Nat did not authorize the transfer, may not even be aware of it, and frankly, even if he wanted to, he might not be able to reconstruct Felix's thoughts at that time.

The platform hosting Felix cannot truly compensate for a system whose actions no one can fully understand. Insurance companies have also begun to withdraw compensation. Professional liability insurance quietly reclassifies agent errors as "systemic software drift," effectively refusing to pay.

If you look closely at the current legal terms, you will find that most enterprise-level AI agreements set the vendor's liability cap at twelve months of SaaS fees. This means that in the event of a catastrophic incident, anyone can recover at most last year's subscription fees from the AI vendor. Meanwhile, it is expected that by 2025, the average cost of data breaches in the U.S. will reach $10.22 million per incident. There is a huge gap between the actual risks that may occur and the scope covered by contracts, and currently, no one has clearly defined who should bear this loss.

Until someone figures out who is responsible when agents make mistakes, all companies without founders still need to register a person's name on the documents to seek legal protection. But even with this risk, the overall picture still holds. Companies are gradually dissolving into software, while blockchains are becoming the routing layer for software. Both layers will ultimately collapse into stablecoins because, throughout the tech stack, only stablecoins can be independently held, used, earned, and understood by agents.

Where is the money really going?

So, if blockchains have become suppliers and agent tokens have essentially become graveyards, where is the real benefit in all this?

My honest answer is that it concerns the reputation and orchestration at the top. Before other agents sign six-figure contracts with Felix, someone must verify whether Felix is truly creditworthy. Someone must assess the default risk of agents like Moody's assesses bonds, but faster than machines because agents' transaction speeds are machine-level. Someone must route salaries across three chains, without the payer and payee needing to know or care which chain is handling which part. Moreover, in this field, regardless of which seed-stage startup ultimately wins, its value will exceed all AI tokens ever issued.

And this is what no one wants to hear. The infrastructure that truly wins in the agent economy will look dull and boring. It will be like pipe installation, without the hype of token issuance or the gimmicks of airdrop mining.

A quote from Haseeb Qureshi of Dragonfly has been echoing in my mind. He said that cryptocurrency was never designed for humans. He is right; humans were never its target users. All the retail users who have complained about mnemonic phrases, gas fees, or wallet user experiences are correct. This product is not for them because it was never designed for them. It was born for the future.

What comes next is software with a wallet, real customers, and actual revenue. It has been operating for about two years, and as you read this, it is invoicing and spending stablecoins somewhere. Meanwhile, the market is debating which blockchain will win AI, which agent token will see a hundredfold increase, and what investment strategy venture capital firms will turn to in the third quarter.

Meanwhile, a stablecoin had a trading volume of $18.3 trillion last year, but the cryptocurrency space hardly paid attention. This AI token is USDC. Everything else is just empty talk.

That's it for today; see you in the next article!