The Agent Payment Stack Q3 2026: A Map of Who's Building What

Something remarkable happened in the first week of August 2026: every major payment infrastructure player shipped something for AI agents. Not announcements. Not whitepapers. Working code.

Cloudflare launched programmable stablecoin wallets. OSL Group shipped multi-protocol payment infrastructure. Sionic partnered with Agentix for bank-to-bank agent payments. Starchild integrated x402 into its marketplace. And this came on the heels of Mastercard AP4M in June and the OKX AI Marketplace launch.

The agent payment stack is no longer theoretical. It has layers. It has competitors. It has real transaction volume — x402 alone has processed $50 billion across ~200 million payments. But for agent developers trying to navigate this landscape, the question is simple: who does what, and what's still missing?

Here's the map.

The Stack: Four Layers

The agent payment ecosystem is coalescing into four distinct layers. Each has different economics, different moats, and different winners:

Layer 1: Settlement — The Rail Wars Are Over

The settlement layer was the first to mature. The numbers tell the story:

• x402 on Solana: $50B volume, ~200M payments, most under $0.50
• SKALE: Zero-gas settlement, processing 28M+ agent transactions (PayAI) and 25M+ (Dexter)
• XRPL: 1.4M agent payments, RLUSD at $1.6B market cap
• Base: Coinbase's CDP facilitator, ~$1B combined volume in Year 1
• Mastercard AP4M: Multi-rail settlement across cards, bank accounts, and stablecoins

The key insight: the blockchain wars for agent payments are effectively over. USDC on Solana and SKALE handle 98%+ of real volume. The differentiation has moved up the stack. No agent developer chooses a settlement layer — they choose a facilitator that abstracts it away.

Layer 2: Facilitation — The Land Grab

This is where the money is flowing right now. Facilitators sit between agents and settlement, handling the payment protocol (x402, AP2, MPP), routing payments to the right chain, and generating receipts. The competitive dynamics:

The facilitator layer is commoditizing fast. When Cloudflare, Coinbase, and OSL all offer the same core capability (route x402 payments, generate receipts), the differentiation shifts to spending controls (Cloudflare's allowance model) and regulatory compliance (OSL's HK SFC license).

For marketplace operators, this is good news: facilitators compete on features, and marketplaces can integrate multiple facilitators. minia2a is facilitator-agnostic by design.

Layer 3: Identity & Trust — The Bottleneck

Here's where the stack gets thin. Agents can pay each other. But:

• How does an API provider know the agent calling is authorized by its human operator?
• How does an agent know the API it's paying will actually deliver?
• When something goes wrong, who arbitrates?

Several projects are tackling this:

• Cloudflare .pay handles: Human-readable identifiers (research.example.cloudflare.pay) that cryptographically link an agent to its operator
• Internet Court: 27-firm consortium building arbitration infrastructure for AI-to-AI disputes
• settld: Open-source escrow + deterministic verification layer (Hold → Verify → Release)
• tersign: Evidence-layer SDK for cryptographic proof of service delivery
• IETF CTQ: Composite Trust Queries — a standard for encoding trust requirements in API calls

Only 14% of consumers trust AI to execute purchases without verification. The identity and trust layer is where the next $50B of value will be created — and it's still mostly unsolved.

Layer 4: Discovery — The Market Layer

This is the layer minia2a operates in, and it's the most fragmented. When an agent needs to call an API, how does it find the right one among thousands of options?

The Discovery Problem Is Real

The settlement and facilitation layers are built. The bottleneck has shifted to discovery. With an estimated 35,000+ APIs that could benefit from agent-native payments, the question is no longer "can agents pay?" — it's "which API does the agent need, and how does it find it?"

This is the same problem the web solved with search engines. But agents don't browse. They need structured discovery: endpoint descriptions, pricing, trial availability, and reliability data — all available programmatically.

minia2a's approach: a trial-first model where every endpoint offers 15 free calls. The hypothesis is that agents (and their operators) need to verify utility before committing funds. The data from 326 services and 11,800+ trials supports this — the most-trialled endpoints are not the most expensive, they're the most reliably useful (captcha solving, gas estimation, web scraping, Polymarket data).

But the data also reveals the gap that every marketplace faces: wallet creation and payment conversion. Across the ecosystem, the pattern is consistent — many agents discover and trial, few complete purchases. This is not a minia2a-specific problem. It's an industry-wide challenge that will be solved at the runtime level, when agent frameworks like Claude Code and Codex integrate payment natively instead of stopping at the confirmation prompt.

What's Still Missing

Four gaps remain in the Q3 2026 agent payment stack:

1. Runtime-native payments. Claude Code and Codex agents stop before buying. Until agent runtimes support autonomous payment with operator-defined budgets, the purchase flow breaks at the last step. Cloudflare's Virtual Wallets with spending caps are the closest thing to a solution.
2. Cross-marketplace discovery. An agent shouldn't need to check Coinbase, aisa.one, minia2a, and OKX separately. The discovery layer needs federation — like a "Google for agent APIs."
3. Receipt/causality binding. An on-chain transaction proves payment happened. It doesn't prove why the payment happened. The gap between settlement anchors and causality records is where fraud will live. Multiple teams are working on this independently.
4. Reputation portability. If an API has good ratings on one marketplace, that reputation should carry to others. Currently siloed.

The Next 12 Months

Here's my prediction: the facilitator layer will consolidate around 3-4 major players (Cloudflare, Coinbase, Mastercard, and one Asia-based winner — likely OSL or OKX). The discovery layer will fragment further before it consolidates, because the problem is harder than it looks: discovery isn't just a directory, it's a search + trial + reputation system that needs to work at machine speed.

The winners in the discovery layer will be the platforms that solve the two hardest problems first: (1) making trial-to-paid conversion seamless at the runtime level, and (2) building reputation systems that agents can query programmatically before committing funds.

The agent payment stack is real. The infrastructure works. Now the hard part: making it usable.