Payments must be reborn for the agentic economy.

The internet was not designed for commerce. This sounds obvious in retrospect, but the consequences are just beginning to surface.

The original protocols — HTTP, DNS, TCP — were optimized for moving documents between researchers at universities. They were never designed to handle value transfer, identity, or trust. So when commerce arrived in the 1990s, we did not redesign the internet. We bolted the existing payment system on top of it: card networks, designed in 1958 for in-person retail, plus a tangle of issuing banks, acquiring banks, gateways, and processors stacked on top.

That bolted-on system mostly worked. For thirty years it powered hundreds of trillions of dollars of online commerce. Stripe became one of the most valuable private companies in the world by polishing it. PayPal scaled it. Square brought it to small merchants.

But the entire system rests on one assumption: that a human being is present, holding a card, deciding to buy.

That assumption is breaking. Right now. Faster than anyone in payments has accepted.

At Sendin, we believe the next decade of commerce will be the largest reshaping of payment infrastructure since the credit card. And the companies that build for the new buyer will define the next century of finance.

#01 · The InheritanceThe internet inherited its payment system from a world that no longer exists.

In 1958, Bank of America mailed 60,000 unsolicited credit cards to residents of Fresno, California.^[1] It was the launch of BankAmericard — what would become Visa. The product was beautifully matched to its era: face-to-face retail, with a clerk verifying signatures, and a 3-day clearing window in mainframe-era banking.

That product is what powers your e-commerce checkout today. The signatures became digital, the clerks became fraud detection algorithms, the mainframes became distributed databases — but the underlying assumption never changed. A human still has to be there.

Every checkout button you have ever clicked exists to satisfy that assumption. Every CVV field. Every 3D Secure popup. Every "Verify it's you" SMS. They are not features. They are friction designed to prove a human is present.

For the human-driven internet, this worked. Slow but acceptable. Expensive but tolerable. The friction taxed every transaction by 2-4%, locked capital in T+2 settlement, and made microtransactions economically impossible. We adapted: subscriptions instead of pay-per-use, cart abandonment as a fact of life, fraud as overhead.

It was a bad system. But it was the only system. And it kept working — until very recently.

#02 · The New BuyerThe world's largest buyer is no longer human.

In November 2022, ChatGPT launched. By April 2024, AI agents were placing real orders. By 2025, the major AI labs — OpenAI, Anthropic, Google, Perplexity — were all racing to ship checkout-capable agents.^[3] The shift is not theoretical. It is happening in production, right now.

Consider what this means. An autonomous software process, with no eyes, no fingers, no human present, is asking your e-commerce site for a product. It cannot fill out your CVV field. It cannot pass your 3D Secure. It cannot click your "I'm not a robot" checkbox — because it is, in fact, a robot.

The card networks have a name for transactions where no human is present at checkout: "card-not-present" fraud. Their entire risk infrastructure is designed to prevent what is about to become the dominant mode of commerce.

By 2030, AI agents will initiate $1.7 trillion in commerce — a 67% compound annual growth rate from $135B today.^[2] Source: Edgar Dunn & Co · 2025

This is not a niche. This is not a future trend. This is the largest wave of new buyers entering the global economy since the rise of the consumer middle class. And they cannot use the existing payment rails.

We have seen this movie before. When mobile commerce arrived in 2008, the industry tried to retrofit cards into apps. It mostly worked, but the experience was so bad it created room for new players: PayPal, Venmo, Apple Pay, Stripe. Most of the value of mobile commerce was captured by companies that built for the new context, not the ones that retrofit the old one.

Agentic commerce is mobile commerce, again, but bigger. And this time, the gap between the new context and the old infrastructure is not 10× wider. It is 1,000× wider.

#03 · The TaxEvery dollar an agent spends pays a tax that should not exist.

Card networks charge 2.5%-3.5% per transaction.^[4] This is not a market rate. It is a structural rent, extracted by intermediaries that the original 1958 design required: issuing banks, acquiring banks, networks (Visa, Mastercard, Amex), gateways, processors, fraud vendors. Each takes a slice. Together they take a tax that flows directly out of merchant margins and into legacy financial infrastructure.

For a human checkout, you might shrug. 3.5% on a $50 sweater is annoying but bearable. But scale it: a marketplace doing $100M GMV gives away $3.5M every year to a payment system designed for an era that no longer exists.

Now imagine the same calculation for AI commerce — where transactions can be sub-dollar, where settlement happens in milliseconds, where agents may execute thousands of micro-purchases per minute. 3.5% is not a tax. It is a wall. It makes the most interesting agent use cases — granular pay-per-use, real-time API consumption, autonomous market-making — economically impossible.

The math is brutal. The card network model breaks at the exact scale where AI commerce becomes most interesting. We are in the early innings of a $1.7 trillion market and the payment layer cannot serve it.

Someone is going to build the new layer. The only question is whether the new layer will be open — owned by no one, available to all — or whether it will be a new generation of walled gardens, where each AI lab tries to lock its agents to its own payment rails.

#04 · The Open Path"Open" is not a marketing word. It is the only path that does not destroy itself.

When the web emerged in 1991, there was no shortage of proprietary alternatives: AOL, CompuServe, Prodigy, Microsoft Network. They had funding, distribution, and superior UX. They lost. They lost because the open web — HTTP, HTML, URLs — created a permissionless surface where anyone could build, link, and compose without asking for access.

The same dynamic will play out in agentic commerce. Today there are early proprietary attempts: Stripe building agent SDKs into its existing rails. Visa announcing "agent payment" pilots. Mastercard demoing agent checkout. Each of these is a closed protocol, owned by an incumbent, designed to keep the existing 3.5% extraction running on top of new technology.

And then there is the open path. Two protocols matter:

• x402 — A revival of the dormant HTTP 402 status code (Payment Required), designed by Coinbase, that lets servers request payment before serving a resource. Open, simple, composable.^[5]
• ACP (Agent Commerce Protocol) — A standard developed by OpenAI, Stripe, and others to let agents negotiate, authorize, and execute purchases programmatically.^[6]

These protocols are not products. They are primitives. Like HTTP, they define how machines speak about value, but they don't deliver the production infrastructure to make it real at scale: merchant onboarding, fraud protection, compliance, settlement guarantees, multi-currency support, dispute resolution.

That layer — the production infrastructure built on top of open protocols — does not exist yet.

#05 · What We Are BuildingThe financial layer the agentic economy was waiting for.

At Sendin, we are building the production-grade payment infrastructure for agentic commerce — the layer that connects open protocols to real merchants, real users, and real compliance.

The thesis underneath every decision we make:

1. Agent-native, not human-retrofit.

Every primitive we ship — wallets, approvals, settlement, authentication — is designed for autonomous machines as the first-class user. We are not adding "AI mode" to a payment system designed for humans. We are rebuilding the stack so agents are the intended species of buyer, with humans as one of many participants.

2. Machine speed, or no speed.

Agents do not wait for T+2 clearing. Sendin settles in seconds, on stablecoin rails. This unlocks the use cases the existing system structurally cannot serve: pay-per-API-call, autonomous market-making, real-time supplier negotiation, granular content monetization.

3. Open standards, not walled gardens.

We build on x402, ACP, and what comes next. We do not own the protocols. We do not gate access by who you are or which AI lab built your agent. We are building the default infrastructure for agents from any vendor to transact with merchants on any platform.

4. Fair economics for every participant.

Our fee is 0.87% — not as a promotional rate, but structurally. We pass the savings of disintermediating card networks back to merchants and users. Sendin makes money by serving more merchants better, not by extracting more from each one.

The window for building this is now. The card networks will not voluntarily give up their tax. The AI labs will keep trying to lock agents to their own payment rails. The existing players will all claim to "support agents" while running 3.5% fees underneath. If the open layer is not built fast and well, the agentic economy will be born locked.

That is what we are at Sendin to prevent. And that is why we are building the financial layer the agentic economy was waiting for — open, instant, fair to every participant.

Human or machine.

ReferencesWhere this thesis comes from.

1. Bank of America · BankAmericard Launch (1958). The Fresno Drop. Visa Inc. corporate history records the distribution of 60,000 unsolicited cards to residents of Fresno, California in September 1958, which became the foundation for today's Visa network. See: Visa — History of Visa.
2. Edgar Dunn & Company · Agentic Commerce Forecast (2025). Projects global AI agent-initiated transaction volume will reach $1.7 trillion by 2030, up from $135 billion in 2024 (67% CAGR). See: Edgar Dunn & Co — Payment Industry Insights.
3. AI agent checkout timeline. OpenAI launched ChatGPT on November 30, 2022. By April 2024, autonomous agents (ChatGPT with browsing, Anthropic Claude tool use, Perplexity Pro Shopping) began placing real e-commerce transactions on behalf of users. See: OpenAI announcements and Anthropic news.
4. Card network interchange rates. Merchant discount rates typically range 2.5%–3.5% for card-not-present e-commerce transactions, composed of interchange (paid to issuing bank, ~1.5%–2%), scheme fees (paid to Visa/Mastercard/Amex, ~0.1%–0.4%), and acquirer markup (~0.5%–1%). See: Visa Interchange Reimbursement Fees and Mastercard Interchange Rates.
5. x402 Protocol. Open protocol reviving the HTTP 402 "Payment Required" status code for agent-initiated programmatic payments. Developed by Coinbase Developer Platform, 2025. See: github.com/coinbase/x402.
6. ACP · Agent Commerce Protocol. Emerging standard for agent-to-merchant commerce negotiations, with contributions from Stripe, OpenAI, and the broader AI-commerce ecosystem. Specifications evolving as of 2026. See: Stripe Newsroom.

Last updated: April 20, 2026 · Find an error? Let us know.