Thirty Years of Ciphers: The Battle for AI Payment Standards
TechFlow Selected TechFlow Selected
Thirty Years of Ciphers: The Battle for AI Payment Standards
A protocol standards battle over “how AI agents autonomously spend money” has become inevitable.
Navigating Web3 tides with focused insightsBy Hazel Hu
AI-powered payments have recently exploded onto the scene—ushering in an inevitable standards war over “how AI Agents autonomously spend money.”
In 1996, Tim Berners-Lee and his team at CERN finalized the HTTP/1.0 specification—and left one slot open in the status code list: 402 Payment Required.
This was a forward-looking interface reserved for future digital cash or micropayment schemes. The architects of the internet foresaw, three decades ago, that the information superhighway would eventually need a toll lane.
Yet 402 has never been activated. No browser supports it; no server truly implements it. It sits like a letter addressed but never mailed—dormant in the corner of the HTTP protocol for thirty years.
Until AI Agents arrived.
During the third week of March 2026, the payments industry dropped four major announcements within 72 hours. On March 17, Mastercard acquired stablecoin infrastructure firm BVNK for $1.8 billion. On March 18, Stripe and Paradigm’s incubated L1 chain Tempo officially launched its mainnet—and simultaneously unveiled the Machine Payments Protocol (MPP). On the same day, Visa Crypto Labs released visa-cli, a command-line tool enabling AI Agents to swipe cards directly from the terminal. Visa’s Head of Crypto, Cuy Sheffield, dubbed it “command-line commerce.” And prior to all this, Coinbase and Cloudflare’s jointly launched x402 Protocol had already been running for nearly a year.
Four moves—on the same board. A long-simmering standards war has officially begun.
The story begins with 402. Marc Andreessen once called the internet’s lack of native payment capability “the original sin of the internet.” HTTP was born to transmit text, images, and video—but not money. This gap gave rise to the entire advertising economy: since users refused to pay for content, advertisers stepped in. And so we got a free—but surveilled—internet.
402 could have changed everything. Imagine your browser requesting a paywalled article; the server responds with HTTP 402 and triggers an in-browser micropayment interface. You click “confirm,” a penny is transferred, and the article unlocks—instantly, as seamlessly as loading an image.
But 1990s technology couldn’t deliver on that vision. There was no mature digital cash solution, no low-cost settlement infrastructure, and SSL/TLS was still in its infancy. So 402 became a placeholder—a standard ahead of its time.
Thirty years later, it was awakened by a Canadian.
Erik Reppel, a computer science graduate of the University of Victoria, served as Engineering Lead at Web3 social startup Zora before joining Coinbase’s Developer Platform. There, he began what he himself calls “work Coinbase has explored since 2015”: designing a native internet payment standard.
One key inspiration came from Balaji Srinivasan’s Bitcoin payment channel experiments at 21.co—when he was Coinbase’s former CTO. That approach was conceptually sound—but failed on cost. Back then, each on-chain transaction incurred gas fees of several dollars, rendering micropayments impractical. By 2024, however, Layer 2 networks had reduced per-transaction costs to under $0.0001.
The timing was finally right. In May 2025, Reppel and colleagues Nemil Dalal and Dan Kim jointly published the x402 whitepaper. Its logic is elegantly simple: a client requests a resource; the server replies with HTTP 402 plus a JSON payload specifying price, accepted tokens, and wallet address; the client signs and pays via wallet; settlement occurs on-chain; the server delivers the resource—all within two seconds.
x402 chose the purest path: protocol-native, on-chain settlement, zero intermediaries. It embeds directly into the HTTP layer—like DNS and TLS—and aims to become foundational internet infrastructure. Coinbase and Cloudflare soon co-founded the x402 Foundation, placing governance in the hands of a neutral entity. Cloudflare’s involvement is pivotal: roughly 20% of global web traffic flows through its 300+ data centers—and x402 is now integrated into its Agents SDK and MCP servers. That distribution advantage is nearly impossible for competitors to replicate.
But another Canadian saw things differently.
Liam Horne, a University of Waterloo graduate, entered the Ethereum ecosystem in 2016—and was a classmate of Vitalik Buterin’s (both also Thiel Fellows). His initial attraction to blockchain was simple: two software systems could transfer value directly—no bank required.
But Ethereum was too slow—and too expensive. In 2017, Horne and Jeff Coleman at L4 Ventures proposed the concept of “generalized state channels”: two parties lock funds into a multisig contract, exchange signed transactions off-chain at arbitrary speed, and only settle the net result on-chain. Zero fees, instant finality, theoretically thousands of TPS.
They also invented a technique called “counterfactual instantiation”—interacting with contracts as if deployed, even though they’d never touched the chain. This idea directly inspired Vitalik’s EIP-1014—the foundation of today’s widely used CREATE2 opcode for account abstraction and cross-chain deployments.
Horne’s most celebrated demo was Web3Torrent: a torrent client where downloaders and uploaders paid micro-fees per file chunk—machine-to-machine, zero human involvement.
But the problem was—nobody needed it.
From 2017 to 2020, DeFi and trading dominated Ethereum’s narrative—not payments. State channels were technically viable, yet found no product-market fit. Like a completed toll booth on an empty highway. Horne admitted state channels became “a solution in search of a problem.” Rollups emerged as the more general scalability answer—and he pivoted to Optimism.
At Optimism, Horne rose rapidly: Engineering Lead, then CEO—growing the team from 8 to nearly 70, leading OP Stack development, and driving Superchain’s collaboration with Base. And Base? Coinbase’s L2 chain. Two threads converged here for the first time.
Later, Horne left OP Labs to serve as advisor—and helped build World Chain. In summer 2024, he dove deep into stablecoins. Base’s success clarified a key insight: the synergy among Coinbase, Circle, and Base closely mirrored Tether, Tron, and Binance’s stablecoin expansion outside the U.S. Stablecoins were no longer niche—they accounted for 70% of all blockchain transaction volume.
In September 2025, Tempo raised $500 million in Series A funding at a $5 billion valuation—jointly incubated by Stripe and Paradigm. Horne joined as Engineering Lead. Paradigm’s CTO Georgios Konstantopoulos and co-founder Matt Huang were deeply involved.
Horne spent six years scaling Ethereum—first via state channels, then Optimism. Then he left—to build a corporate-backed, non-decentralized L1. In the same week Tempo launched its mainnet, OP Labs announced a 20% layoff—and Vitalik had effectively publicly dismissed the L2 roadmap. His classmate and Thiel Fellow seemed to have made his choice earlier.
This intriguing personnel shift wasn’t isolated.
In February 2026, Dan Romero and Varun Srinivasan—co-founders of Farcaster, the decentralized social app backed by Vitalik—announced they were joining Tempo. Before founding Farcaster, Romero was Coinbase’s 20th employee (joined 2014), rising to VP overseeing international and consumer business. Srinivasan also came from Coinbase, where he led engineering and product.
Their exit from Farcaster was dramatic: after Neynar acquired the Farcaster protocol, the founding team collectively departed—and Romero announced he’d return the full $180 million raised to investors. He later tweeted: “Stablecoins are a generational opportunity.”
It doesn’t stop there. Tempo also recruited Dankrad Feist—one of Ethereum Foundation’s key L1 scaling architects—meaning Tempo’s core team now includes at least two former Coinbase executives, a former Ethereum L2 CEO, and a former Ethereum Foundation researcher. Meanwhile, on the other side, Coinbase is pushing x402 full throttle. In a sense, this standards war is both a showdown among Coinbase alumni—and a quiet schism within the Ethereum ecosystem.
So what exactly does Tempo’s MPP do—and how does it differ from x402?
Horne designed MPP around one core idea: “state channel sessions, lifted to the application layer.”
In x402, every API call by an Agent triggers a separate on-chain transaction. Fine for low-frequency use—but when an AI Agent makes thousands of calls per minute, each requiring on-chain settlement, latency and friction skyrocket.
MPP solves this with Sessions: the Agent performs one handshake, sets a spending cap, then freely invokes hundreds or thousands of APIs within that Session—all charges settled in a single batch upon Session close. Structurally, it mirrors Horne’s nine-year-old Ethereum state channel work: fund locking, off-chain interaction, one-time settlement.
Another critical difference lies in payment methods. x402 uses purely on-chain stablecoin settlement—naturally decentralized, yet inherently incompatible with fiat. MPP adopts a “payment-rail-agnostic” approach, supporting stablecoins, credit cards, and even Bitcoin’s Lightning Network. Lightning integration is handled by Lightspark—whose founder is none other than David Marcus, former PayPay president and core architect of Libra.
x402 also introduces a key role: the Facilitator. Neither buyer nor seller, it acts as a trusted third-party verifier. After the Agent signs a payment, funds don’t go directly to the merchant—but first pass through the Facilitator, which validates signatures, checks balances, executes on-chain settlement, and only then releases the resource. An extra hop—but one that shoulders all heavy lifting: anti-fraud, compliance checks, and settlement execution. Merchants need only know: “The Facilitator says OK—payment received.”
MPP eliminates this role entirely. Horne’s rationale: if both Agents and merchants grow increasingly intelligent, why retain a middleman? Verification, encryption, and settlement are all handled directly by the merchant’s server. Shorter chains, lower latency, no intermediary markups.
But the trade-off is direct. Card encryption/decryption logic falls squarely on merchants—and PCI compliance responsibility shifts with it. For large merchants already embedded in Stripe’s full stack—Shopify, DoorDash—this isn’t a hurdle. But what about tomorrow’s merchant: an AI Agent itself—a “zero-human company” with no compliance team, no legal department? How will it handle these requirements? MPP offers no answer—yet.
Visa has also authored card-payment specifications for MPP. For merchants, integrating MPP means plugging into Stripe’s existing infrastructure—tax calculation, fraud protection, refunds—all pre-integrated and ready to deploy. Stripe’s cold-start advantage across millions of global merchants is something x402 cannot match in the short term.
Since launching in May 2025, x402 claims to have processed over 100 million payments. Bloomberg cited x402.org data reporting $24 million in Agent payments over 30 days. Yet raw on-chain analytics from Allium Labs show only ~$3 million in actual on-chain volume during that period. Artemis Analytics analysts applied wash-trading filters—flagging wallets repeatedly transacting with themselves or cycling funds between addresses—and arrived at a final figure: $1.6 million.
This gap doesn’t necessarily signal x402’s failure—but it starkly reveals how early the Agent payment space remains. What are real Agents actually buying? Almost exclusively developer tools billed per use: Firecrawl charges one cent per web scrape; Browserbase sells browser sessions; Freepik sells AI-generated images. Users aren’t drawn by crypto—they’re drawn by “no sign-up, no subscription, just sign with your wallet and go.” The author of that data analysis piece summed it up: he spent $0.47 total using x402 to fetch data from Allium and write the full article.
$1.6 million isn’t a large number. But Stripe, Cloudflare, and Google aren’t betting on a $1.6-million-per-month market. They’re betting on the moment Agents become default buyers—and the scale that follows. McKinsey projects Agent payments will reach $3–5 trillion annually by 2030.
Interestingly, the market hasn’t picked sides.
Nearly every major player is hedging its bets. Stripe itself co-authored MPP—and was also an early x402 integrator. Anthropic supports both protocols via MCP. OpenAI runs demos on MPP—and appears in x402’s ecosystem. Google hasn’t built its own settlement protocol, but created an authorization framework, AP2—and embedded x402 directly inside it. This shows the conflict runs deeper than “two protocols fighting.”
It’s rational hedging—akin to early internet companies supporting both HTTP and HTTPS—not because they couldn’t tell them apart, but because nobody wanted to bet wrong during migration.
Yet the underlying philosophical divide is real.
x402 believes in “protocol-as-infrastructure.” It aims to be the TCP/IP of payments: permissionless, chain-agnostic, universally accessible—no reliance on any commercial entity. Ethereum, Base, and Solana ecosystems are fully embracing x402—it’s becoming the core payment rail for Agent-to-Agent economies.
MPP believes in “infrastructure-as-protocol.” Backed by Stripe and Visa’s commercial empire, it first captures Agent payment demand within the traditional payments world—then gradually extends on-chain. Merchants already accept payments via Stripe; adding MPP lets them instantly accept automatic Agent payments.
Some compare them to TCP/IP vs. AOL in the internet’s early days. But that analogy isn’t quite right—MPP is open-source and claims to be an open standard. As multiple developers point out: MPP’s documentation explicitly recommends Tempo as its settlement layer, and its SDK currently supports only TypeScript—while x402 already offers multi-language support (TS, Python, Go). One assessment cuts straight to the point: MPP is “a vertically integrated, non-neutral version of x402.”
But regardless of which protocol wins, both require a blockchain foundation—yet stablecoins’ true base today lies on no new chain.
Ethereum mainnet plus its rollups (Arbitrum, Base, Optimism) hold over $170 billion in stablecoin supply—with ~$2.8 trillion in monthly transaction volume. x402 currently runs primarily on Base and Solana—the real battleground for Agent micropayments. Dig deeper: Tron processes over $600 billion in stablecoin transfers monthly—almost entirely sub-$1,000 cross-border remittances. It’s absent from Silicon Valley’s narrative—but it’s the world’s most-used stablecoin network.
Onto this established landscape, two new “stablecoin-first” chains have just entered the fray. Stripe has Tempo; Circle has Arc. Both target sub-second finality, predictable fees, and compliant privacy—and both claim to be next-generation payment infrastructure. Yet neither has undergone large-scale real-world validation: Tempo launched its mainnet on March 18; Arc remains on testnet, targeting mainnet launch in 2026.
Two new chains challenge three legacy chains and their L2s; two protocols battle for standards dominance; multiple payment giants jockey for position. The table grows ever more crowded.
Yet at least, Agent Payments has brought fierce new challengers to the field. Ethereum and Solana were built for general-purpose computation—stablecoin payments run fine on them, but aren’t optimal. Gas fees fluctuate with network congestion; settlement certainty depends on overall chain load; compliance and privacy require custom layers. Agents demand certainty: fees predictable to the cent, finality measured in milliseconds, compliance frameworks out-of-the-box. These three new chains were architected from line one for those exact needs—betting that the Agent payment growth market is large enough to bypass legacy chains’ incumbency advantages—and establish sovereign ground in this new arena.
The wager’s logic is simple: Agent count will likely dwarf human count. A person swipes a card a few times per day; an Agent may invoke hundreds of APIs per second—each call a micropayment. Coinbase’s Brian Armstrong predicted online Agents will soon outnumber human users. Yet today’s Agents hold virtually no wealth—they lack wallets, independent budgets, and must seek human approval for every cent spent. This standards war is ultimately about capturing the inflection point where Agents evolve from “running errands for humans” to “spending autonomously.” Whose protocol has the pipes laid down first—that’s who catches the wave.
Tim Berners-Lee in 1996 couldn’t foresee this. He left a payment interface in HTTP—as if carving a window into a blank wall. For thirty years, that window stayed shut—until AI Agents poked their heads in and said: “I need to spend money.”
Multiple teams rushed in—competing to frame that window. One believes in open protocols; another leans on commercial empires; others race to build trust layers between them. Many standing in the middle share names—and even offices—before walking away in opposite directions.
This standards war has only just begun. Victory may not hinge on technical superiority—but on who first makes that window indispensable. Because internet history teaches us: it’s never the most elegant protocol that wins—it’s the one most people use.
But one thing is certain: that dormant 402 status code—sleeping for thirty years—has finally awakened. And this time, the one knocking isn’t human.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
@PodOur2Cents
