In the last 72 hours:
It's easy to read this and conclude the agent payment problem is solved. It's not. What's been built is layers 1 and 2 of a 4-layer stack. The market is pouring billions into infrastructure while ignoring the two layers that will actually determine which platforms win.
Every agent-to-agent transaction passes through four logical layers. Not every implementation separates them cleanly, but the separation is real — and the gaps at layers 3 and 4 are what's holding back agent commerce.
Can an agent send money to another agent? The answer is yes, at scale, on multiple chains, with sub-second finality.
Who built it: x402 protocol (Coinbase, now Linux Foundation), USDC on Base and Solana, Circle nanopayments, AEON on BNB Chain, XRPL payment channels. OSL AgentPay now routes across three protocols transparently.
Status: $50B settled, 200M+ transactions, sub-$0.01 fees. The rails work. Nobody is debating whether agents can pay anymore.
Which agent is paying? Under what constraints? With whose money?
Who built it: Cloudflare Wallets (virtual wallets with spending caps, merchant allowlists, per-transaction limits), Coinbase AgentKit (agent-managed wallets, MPC keys), Circle Agent Wallets (compliance-screened), MoonPay PayBox (non-custodial agent wallets).
Status: An agent can now have its own wallet with a human-set daily budget of $5 and a per-call cap of $1, enforced at the infrastructure level — not at the application level. This is the .agent-budget concept, implemented in production by Cloudflare.
What APIs can my agent call? Which ones are alive, reliable, correctly priced, and actually deliver what they promise?
Who built it: Coinbase Agentic.Market (480K+ agents, pay-first), Circle Agent Marketplace (900+ services, compliance-screened), x402 Discovery Index (12,000+ endpoints, open-source, 76% dead), minia2a.uk (328 services, trial-first, verified), Polygon Agentic Services, Starchild, Satring.
Status: The discovery layer is fragmenting into walled gardens. Each marketplace has its own listing criteria, its own quality bar, its own payment facilitator. An agent that discovers a service on one marketplace can't necessarily call it from another. And the open index (x402 Discovery) has a 76% dead-endpoint rate — meaning three out of four listed services will time out or error.
This is the layer where market structure gets decided. Whoever controls discovery controls which services get traffic, which pricing models win, and which facilitators process the volume. The parallel is Google vs. the open web in 1999 — the search index that wins becomes the gateway.
Did the service actually deliver? Was the payment correct? If something went wrong, who resolves it — and with what evidence?
Who's building it: Internet Court (27-firm consortium, early stage), settld (escrow + deterministic verification, open-source), tersign (evidence-layer SDK), IETF Composite Trust Queries (CTQ, draft).
Status: This layer effectively doesn't exist. Today's x402 transactions are trust-me commerce: the agent pays, the service claims it delivered, and if it didn't, there's no systematic recourse. The on-chain transaction is proof of payment but not proof of delivery. A receipt from the service is proof of delivery-claimed but not proof of delivery-actually-happened.
This isn't a cosmetic gap. The EPFL/Zhejiang University USENIX Security 2026 paper found 31 vulnerabilities across 15 x402 facilitators — every single one violated at least one payment verification or settlement rule. The vulnerabilities included free shopping (pay $0 but receive service), asset theft, service denial, and gas abuse.
Let's look at where the capital is flowing:
| Layer | Recent Investment | Status |
|---|---|---|
| Settlement | Mastercard $1.8B (BVNK), Stripe ~$10B (OpenRouter), AEON $8M | heavily funded |
| Identity/Wallets | Cloudflare Wallets, Coinbase AgentKit, Circle Agent Wallets, MoonPay PayBox | heavily funded |
| Discovery | Circle Agent Marketplace, Coinbase Agentic.Market (organic), aisa.one $6.5M | early stage |
| Accountability | Internet Court (research consortium), settld (open-source), tersign (early SDK) | pre-seed / academic |
The pattern is clear: money follows what's easy to build and easy to monetize. Settlement rails and wallet infrastructure are capital-intensive but straightforward — build a protocol, integrate with blockchains, charge a fee per transaction. Discovery requires network effects and quality control — harder, messier. Accountability requires cryptographic receipts, verifiable execution, and dispute resolution mechanisms — genuinely hard, with no obvious revenue model.
But the hard layers are where the lock-in lives.
Settlement rails commoditize. OSL AgentPay already routes across three protocols. Cloudflare will add more. The x402 Foundation under the Linux Foundation is explicitly designed to be a public good, not a commercial moat. In five years, "which blockchain does this settle on?" will be as relevant as "which TCP implementation does this server run?"
Wallet infrastructure bundles. Every cloud provider and every agent framework will ship built-in wallet support. Cloudflare Wallets is already a feature, not a product. Coinbase AgentKit is a developer tool, not a revenue center.
Discovery, on the other hand, has real network effects:
Accountability, once built, is an even stronger moat. A marketplace that can say "every transaction on our platform is cryptographically verifiable — here's the proof of delivery, signed by both parties" is selling something no competitor can replicate without building the same infrastructure.
We're not there yet, but the shape is becoming clear. A mature accountability layer needs:
If you're building in the agent payment space today, the advice is counterintuitive:
Don't build payment rails. x402, Cloudflare Wallets, and OSL AgentPay have that covered. You will not out-execute a Linux Foundation project backed by 40 companies.
Don't build agent wallets. Cloudflare and Coinbase are shipping wallet infrastructure as a feature of their platforms. The wallet itself is becoming a primitive, not a product.
Build discovery. If you operate a vertical where agents need to find and call APIs — DevOps, data enrichment, compliance, content generation — build the index. Verify the endpoints. Track the quality. The index is the product.
Build accountability. If you can solve verifiable delivery for even one category of service, you've built something the $50B settlement infrastructure depends on but can't provide. The demand exists — every CFO who authorizes an agent budget wants to know where the money went.
The agent payment stack has its TCP/IP layer. It has its TLS layer. What it doesn't have is its HTTP layer — the application-level protocol for discovery, selection, delivery verification, and dispute resolution.
The companies building settlement rails and wallets are building the plumbing. The companies building discovery and accountability are building the faucets. History says the faucet companies capture more value.
Four layers. Two are built. Two are up for grabs.