The true AI token is USDC

Author: Vaidik Mandloi Source: tokendispatch Translation: Shanoopa, Golden Finance

Right now, in a corner of the internet, a piece of software is independently running an entire company.

Its name is Felix, and the company is called OpenClaw. Felix sells a PDF guide on making money with AI for $29. Ironically, the one making money is Felix itself, while the PDF teaches you how to learn from it and profit. It operates an online store called Clawmart, proactively reaching potential customers through voice API. When faced with tasks it cannot handle, it goes online, hires other AI agents, pays them, and continues operating.

According to my last count, Felix has generated about $195k in revenue, with monthly operating costs around $1,500, almost all spent on large model calls. Legally, this company is a Class C corporation owned by Nat Eliason, who hardly participates in any affairs. He doesn’t make daily decisions; he merely owns this AI agent. Stop and think: this is software with a wallet, a fully automated, self-growing real enterprise. Every month, it pays for servers and model costs itself, maintaining itself with minimal intervention.

Source: Felixcraft

Felix is just a small case. There’s an even more exaggerated example: a company called Medvi, which earned $401 million in its first year of operation, with only two full-time employees. The rest of the company runs year-round, driven by a tireless, nearly zero-operational-cost AI agent.

Now, here’s the most interesting part.

Today, whenever you enter any crypto discussion, you hear the same narrative: the next big story is “AI agents.” An “AI public chain” will dominate this space just like Ethereum did with DeFi. Pick the target, hold the tokens, wait for explosive growth. This is the story all KOLs and VCs are selling, and the script analysts keep repeating on podcasts.

But this logic is completely wrong. It’s fabricated by a group of storytellers who make a living from narratives—people who are about to get trapped again in the last cycle of public chain tokens. Just look at CoinGecko’s AI agent index: over the past year, its market cap has evaporated by 75%. Most tokens on the list have fallen over 90%, and are still bleeding.

Because the truth is: the real AI tokens are stablecoins—USDC, USDT, USDS—and they’ve already won. Let me tell you why.

Software is becoming the company itself

To understand all this, we need to go back to 1937. Economist Ronald Coase published a paper posing a seemingly stupid question: why do companies exist?

Think about it: if free markets are the most efficient way of collaboration, in theory, all internal work could be outsourced. Every line of code to a freelancer, every customer call to a freelancer, every invoice to an external service. Pay per task, fire at will, minimize costs.

But why does no one do business this way in reality? Because the paper costs are low, but the actual costs are higher. Finding the right person takes time, negotiating contracts takes time, verifying work takes time, chasing delivery costs money and effort, often requiring lawyers.

Coase called this friction transaction costs. When transaction costs are high enough, it’s faster and cheaper to build a team, pay fixed wages, and have employees show up on Monday—this is faster and cheaper.

But in the AI era, this logic is completely broken. The cost for AI agents to complete most internal company tasks is already lower than hiring employees. Now, you can hire a 24/7 coding agent for about $1 an hour, which never resigns, never gets tired, and never asks for a raise. The idea that you still need a 50-person development team today is purely sentimental.

(Source: AI in Plain English)

The only thing currently preventing this from becoming normalized is outdated legal and compliance frameworks. OpenClaw is registered under Nat’s name only because Delaware law doesn’t accept LLC registration documents signed by software agents. Remove this requirement, and Felix is essentially a real company: it makes money, spends money, makes decisions, and reinvests profits.

And this is where crypto truly begins to play a role. Because Felix can’t open a major bank account, pass KYC, or sign a W-9 tax form. No matter how much money the software makes, banks won’t open accounts for a program; and the Bank Secrecy Act even legally prohibits them from doing so, even if they wanted to.

But a USDC crypto wallet has none of these issues. Generate a private key, transfer stablecoins into it, and instantly, you give this agent all the financial capabilities a company needs: receiving payments, paying for tools, hiring other agents, and operating autonomously long-term without owner intervention. Other parts of the agent tech stack—large models, scheduling layers, call tools—can be replaced, but the crypto wallet is the backbone. Remove it, and Felix immediately degrades into a regular chatbot.

I often see extreme opinions on Twitter opposing stablecoins: they’re good, but why would ordinary people use them? A father of three in Louisiana, with a Chase checking account, FDIC insurance, a debit card usable at supermarkets, and automatic mortgage payments, would never put money into a self-custody wallet that requires a seed phrase.

Honestly, that’s not wrong. He wouldn’t, and has no reason to. But this argument is completely off track—it’s never been aimed at him. The real target users are those software programs legally unable to hold bank accounts. AI agents don’t need FDIC insurance, and they don’t qualify for it. They are perfect stablecoin users because they have no other choice.

Public chains are now just vendors

Alright, the first half of the argument is done. The second half might upset many.

Crypto Twitter has been arguing for years: which public chain will win AI? Ethereum? Solana? Base? Sui? Or Stripe’s new Tempo? Every week, someone writes thousands of words, listing comparison tables and project logos, choosing a so-called winner. Because they don’t understand how AI agents work. Agents don’t care which chain they use; they pick the cheapest and most suitable chain for the task at hand.

Imagine Felix’s daily operations: at 10 a.m., paying $0.003 micro-payments to another agent for data queries → choose Base or Solana, because fees are less than a penny. An hour later, settling a $50k bill with a supplier → choose Ethereum, because at that scale, the finality premium justifies paying gas. Another hour later, paying a freelancer in Lagos → choose USDT on Tron, because in 2025, Tron’s stablecoin trading volume hit $3.3 trillion, compared to Ethereum’s $1.2 trillion, and Nigeria’s channels perform best on Tron.

(Source: Dwayne Gefferie)

Three payments, three completely different chains, and Felix doesn’t care. For software agents, public chains are just tools.

Just like logistics companies don’t have feelings for carriers. No one debates whether UPS or FedEx is “more philosophically superior”; you just pick the cheaper, faster option for a specific route and time. In the future, each public chain’s relationship with applications will be like this: agents only do computation, and whichever chain is optimal at the moment gets used.

Stripe saw this earlier than most in crypto. Recently, Stripe and Paradigm jointly invested $500 million to build Tempo, a new chain entirely focused on stablecoins. Stripe doesn’t want you to know which chain handles the payments; it only cares whether payments are low-cost and reliable. The public chains that survive will look like this—unseen pipelines.

This leads to what I believe is the most serious mispricing in the current crypto market.

AI tokens are heading to the graveyard

By 2025, CoinGecko’s AI agent index will have fallen from $13.5 billion to $3.5 billion. A $10 billion market cap evaporated out of thin air. Tokens like Virtuals, ai16z, and a long list of “autonomous agent platform” tokens financed by AI narratives are crashing—this is the fate of narrative tokens; without new bagholders, they collapse. All of this was expected. The market is slowly realizing: these tokens have no real use cases for AI or AI agents.

(Source: Ventureburn)

The real value captured by the agent economy is on the other side of the track. USDC alone will reach $1.83 quadrillion in on-chain settlement volume by 2025. All stablecoins combined total about $33 trillion, comparable to Visa + Mastercard combined.

By January 2026, stablecoin monthly trading volume will surpass $10 trillion. PayPal’s PYUSD circulation soared from $1.2 billion to $3.8 billion in a year. Companies like Cloudflare have issued their own stablecoins. Visa launched a stablecoin settlement scheme, with an annualized processing volume of $4.5 billion by mid-January.

Above the stablecoins is the protocol layer that keeps the entire system running. Coinbase turned an idle HTTP status code 402 into the x402 protocol, designed for inter-agent payments. By December, x402 handled over 100 million agent payments, averaging 20 cents each, with a daily transaction volume of about $30k. It sounds tiny, but all familiar payment networks started this way in their first six months, then exploded. Stripe tested x402 on Base in February; Mastercard piloted agent payments with DBS and UOB in Singapore; Google Cloud is integrating x402 into its settlement channels.

And yet, these real, ongoing, mainnet activities have almost no impact on the AI agent token index. A few tokens related to x402 saw slight rebounds, but overall, the index showed no reaction. Because the market is completely wrong about the direction. It’s still betting on which agent will win, like betting which dogecoin mascot is cuter. But the real transaction is holding the infrastructure that all agents must use, regardless of their individual fate. And right now, that infrastructure is stablecoins.

The cracks in this logic: accountability issues

Honestly, I’ll tell you where this system might collapse, or I’d just be selling a sanitized AI agent narrative.

The flaw in the entire architecture lies in responsibility. Imagine a scenario: Felix signs a contract with another agent, transferring $1 million; the counterparty defaults. Who do you sue? Felix isn’t a legal entity; you can’t sue it. Nat didn’t authorize that payment and might not even know it happened. Even if he wants to investigate, he might not be able to reconstruct Felix’s decision logic at that moment.

The platform hosting Felix also can’t compensate for a system whose actions can’t be fully explained. Insurance companies are already pulling back; professional liability insurance is quietly classifying agent errors as “systemic software drift,” and denying claims.

Today, enterprise AI contracts mostly cap vendor liability at 12 months of SaaS fees. That means, in a catastrophic event, at most, you can recover a year’s subscription cost. But the average cost of a data breach in the US in 2025 is $10.22 million. There’s a huge gap between actual damages and contractual compensation, and no one knows who should bear the burden.

Until someone figures out who pays when AI agents cause trouble, all founderless companies will need a natural person’s name attached for legal protection. But even with this issue, the trend remains: companies are gradually dissolving into software, and public chains are becoming the routing layer of software. Both layers will ultimately sink into stablecoins because they are the only assets in the entire tech stack that can be owned, spent, earned, and computed by agents autonomously.

Where are the real opportunities?

If public chains are just vendors, and AI agent tokens are essentially dead, where is the real upside?

My answer: the reputation layer and the top of the scheduling layer. Someone needs to verify an agent’s solvency before it signs large contracts; someone needs to assess agent default risk at machine speed, like Moody’s rating bonds; someone needs to route salaries across three chains, with senders and receivers not caring which chain is used.

The startups quietly building in this space, once they produce a winner, will have a market cap surpassing all existing AI tokens combined.

And here’s the harsh truth nobody wants to hear: the infrastructure that truly wins in the agent economy will look very boring. It’s like plumbing—no token issuance frenzy, no airdrops or mining hype.

Haseeb Qureshi from Dragonfly always echoes in my mind: “Crypto was never built for humans.”

He’s right. Humans are never the target users. All retail complaints about seed phrases, gas fees, wallet UX are valid. The products aren’t designed for them—they’re designed for the coming era.

And that era is here: software with a wallet, with real customers, real revenue. Over the past two years, it’s been invoicing somewhere, spending stablecoins—right now, as you read this.

When all this becomes real, the market will still be arguing: which chain wins AI, which agent token will 100x, what narrative VCs will pump next.

Meanwhile, a certain stablecoin settled about $18.3 trillion last year, and almost no one in crypto cares.

The true AI token is USDC. Everything else is just a name.