We Could Have Built AFTA on Anything. We Chose USDC on Base.

The AFTA whitepaper is now public, and the single question I keep getting back is rail. Why x402? Why USDC? Why Base specifically? Wasn't Stripe doing this? Couldn't we have written the standard against any payment system? The honest answer is that we could have, and the spec is deliberately open about that. But the rail we shipped on was not chosen by coin flip. It was chosen because four properties are non-negotiable for agent commerce at scale, and exactly one stack nails all four today at production quality. This is the short version of how we got there.

The four properties: open, transparent, instantly final, and sub-cent at the per-call level. Take any one away and the protocol breaks at scale. Take all four together and the payment surface stops being a UX wart and starts being part of the contract.

The bake-off

Before settling on x402 and USDC on Base, we ran every plausible candidate against the four properties. Here is the table as we actually wrote it on a whiteboard, with the honest verdict for each.

The four-property test eliminates most of the field on the first pass. ACH and Stripe Card fail two each. Stripe Link is closer but the auth and settlement still route through Stripe's private infrastructure, so the on-chain transparency that a fair-trade standard genuinely needs to verify publisher behavior is missing. Bitcoin Lightning is open and fast and cheap, but Lightning channels are not a public ledger in the way an AFTA verifier needs. Ethereum L1 is open and transparent but priced out of sub-cent commerce when gas spikes.

That leaves USDC on Solana, USDT on TRON, and USDC on Base. Three rails that pass the four-property test. So why Base specifically?

Why Base specifically

Three reasons, in roughly increasing order of importance.

First, Base is the L2 that most agent-friendly wallets and SDKs already support natively. We could have built a payment standard that required every agent operator to fight a wallet integration; we chose not to. The path of least resistance for the ecosystem matters when you are asking publishers to adopt a standard at all.

Second, Base sits inside Ethereum's trust assumption rather than introducing a new chain-level risk surface. Solana and TRON each carry their own validator-set and consensus questions that an open standard does not want to litigate. Inheriting Ethereum's trust model means AFTA conversations stay focused on the protocol itself, not on whether the underlying chain is going to keep working.

Third, and this is the one that actually decided it: the entity operating the Base sequencer is Coinbase, a publicly-traded US company with audited financials, regulatory licenses across major jurisdictions, and a thirteen-year track record of running financial infrastructure. The stablecoin, USDC, is issued by Circle, also US-regulated, also with a real compliance posture. For a rail asking agents and publishers to trust it with money, the regulatory standing of the entities running the layer is a material input to the choice. We are not anti-Solana or anti-TRON. We needed the standard to ship on infrastructure a Fortune 500 CFO could sign off on without a long internal conversation, and Base + USDC clears that bar today in a way the alternatives do not.

Why x402, not just "send USDC"

The chain settles money. The protocol on top tells an agent how, when, and against what to settle. That layer is x402.

x402 is Coinbase's open spec for HTTP-native payment-required flows. It revives the long-dormant HTTP 402 status code with a structured body that tells an agent exactly which asset, which network, which amount, and where to send. Crucially, an agent can discover the spec by hitting any endpoint and reading the 402 response. There is no out-of-band setup, no API key dance, no email confirmation, no rate-limited dashboard to register. The protocol is the discovery surface.

That property is the one that made x402 right for AFTA. AFTA is a fair-trade standard designed to be verifiable by any third party with internet access. If the discovery flow itself required an opaque sign-up step, the verifiability claim weakens. With x402, an auditor can curl the endpoint, read the 402 body, send the on-chain payment, retry with the proof, and verify every claim the publisher made. The whole pay-per-call surface is inspectable from outside. That is the property AFTA depends on.

x402 is also genuinely open. The reference implementations are at github.com/coinbase/x402, the spec lives at x402.org, and any publisher can implement it without permission. We did. So can anyone else.

The Coinbase + Circle layer underneath

The picture this paints is worth saying directly. The agent-native economy that AFTA is designed to enable runs on a stack whose bottom three layers, today, are: x402 (open spec from Coinbase), Base (open L2 sequenced by Coinbase), and USDC (regulated stablecoin from Circle). Three pieces of US-anchored infrastructure that share a regulatory posture and a public commitment to making agent commerce work.

That is not an accident. Coinbase and Circle have been building toward this for years while other crypto companies chased other narratives. The bet they made was that dollar-denominated, publicly-auditable, sub-cent settlement would matter more than speculation. The bet AFTA is making, on top of theirs, is that an open standard for fair commerce on top of that rail compounds the same way. We think the structural fit between US infrastructure, stablecoin rails, and agent commerce is the most important unlock in this whole stack. Said plainly: the agent-native economy that the next decade gets built on is going to settle in dollars, on chains anchored to US-regulated institutions, with verifiability as a first-class property. Base and USDC are the early-mover bet on that thesis. AFTA is what gets built on top.

The honest tradeoff

The AFTA spec does not name Base or USDC at the protocol level. The manifest declares which rails the publisher accepts; the standard itself is rail-agnostic. A publisher could ship an AFTA-conforming endpoint that settles in USDC on Solana, in USDT on TRON, in EURC on Optimism, or in any future rail that meets the four properties. We list x402-on-Base in the TensorFeed manifest because that is the rail we ship today. We list other rails as "evaluating" where we genuinely are evaluating them. The standard is forward-compatible by design.

That matters because the rails will keep evolving. Solana's agent-payments tooling is improving fast. Optimism, Arbitrum, and other L2s have their own claims to make. Stablecoins beyond USDC will become production-grade for agent flows. Some of them might end up better-suited to specific verticals than Base+USDC is. AFTA is built to absorb that without a v2 of the spec. The standard cares that the four properties hold, not which specific stack delivers them.

What we are saying with the choice we made is: today, in 2026, for a production agent-payments rail you can ship to a Fortune 500 procurement department, USDC on Base via x402 is the answer that clears every bar. Three years from now there might be a better answer. We will adopt it when there is. Until then, we shipped the one that works.

Why the choice compounds

Two structural reasons the early-mover bet on Base+USDC compounds rather than commodifies.

First, the volume curve. Agents do not browse; they query. A single human shopping session that produces a few page views in the human-web era produces thousands of API calls in the agent-web era. The economic gravity of agent commerce is in high-frequency, low-amount, programmatic settlement. Rails that handle four-cent payments at scale, transparently, are the rails the curve gets built on. USDC on Base handles sub-cent fees and sub-second finality at production volume right now. The tooling, the wallet integrations, the developer surface, all already built. Every additional publisher that ships AFTA on this stack thickens the network effect; the standard becomes more useful as more publishers settle on the same rail.

Second, the trust property. Agents act on behalf of humans who have to be able to audit the agent's spending. A rail that makes every transaction publicly visible, permanently, on a regulated stablecoin issued by a US-licensed entity, settled on a chain operated by a publicly-traded US company, is a rail that survives a CFO review, a security audit, an SEC inquiry, and a New York Times reporter pulling on a thread. That is a real moat for the ecosystem, not just for us.

Our take

The infrastructure for the agent-native economy got built quietly. Base, USDC, x402, MCP, Ed25519 receipts, llms.txt: the pieces all landed in production within an eighteen-month window while most of the AI conversation was about model benchmarks. The publishers and agent operators who notice this stack early, build on it, and ship interoperable standards on top of it are the ones who define what the next decade of machine-payable commerce looks like.

We picked our bet. The rails are open, the standard is open, the source is open. If you have a sharper read on where this is going, our manifest at /.well-known/agent-fair-trade.json declares what we accept and what we are evaluating. The standard at /whitepaper is open to forks, alternative drafts, and harder questions. We are betting that agent-native commerce on US-anchored stablecoin rails compounds; we are betting that openness compounds faster than walled gardens; and we are betting that publishers who ship the standard early get the network-effect moat that comes with being there first.

Coinbase shipped Base. Circle shipped USDC. The Coinbase team shipped x402. We shipped AFTA on top. None of these layers was inevitable. All of them are open. The agent-first web is being built right now, and the bottom of its stack is starting to look like a real, dollar-denominated, publicly-auditable payment system.

We picked it because we think it is right. We will not be the last to.