The Agent Payment Stack Is Forming — And Discovery Is the Missing Layer
August 9, 2026
In the last three months, the agent payment infrastructure space went from "Coinbase and a few startups" to "every major financial institution has a product." The stack is forming fast — but one layer is conspicuously underbuilt.
AI agents have already processed 176 million transactions settling over $730 million. 76% of those payments are under $0.30 — the threshold where traditional card processing fees make payments economically nonviable. This is why the entire stack runs on stablecoins.
The Three-Layer Stack
What's emerging isn't a single "agent payments" category — it's a stack with distinct layers, each solving a different problem:
Layer 1: Settlement Crowded
How does money actually move from agent A to service B? This is the most competitive layer, with six major entrants in 2026 alone:
| Player | Product | When | Approach |
|---|---|---|---|
| Coinbase | x402 on Base | May 2025 | USDC wallets as API keys; ~3,900 merchants |
| Circle | Agent Stack + Nanopayments | May 2026 | Gas-free USDC, off-chain aggregation, $0.000001 minimum |
| Stripe + Tempo | Machine Payments Protocol | Mar 2026 | Payment-agnostic: cards, crypto, Lightning |
| Mastercard | Agent Pay for AI | Jun 2026 | On-chain permissions on Polygon; Adyen + Coinbase + Cloudflare partners |
| AWS | Bedrock AgentCore Payments | May 2026 | USDC via x402 on Base, ~200ms settlement |
| Solana Foundation | Pay.sh | Early 2026 | Open-source, USDC on Solana, built with Google Cloud |
Also: AutoIncentive (open-source x402 facilitator on SKALE, 1,000+ txns), Swarm on Arc (agentic commerce with bonded reputation).
Layer 2: Authorization Building
Who gave the agent permission to spend, and under what constraints? This is the policy layer — can an agent spend $0.01 on a gas-price lookup but not $5 on an LLM call?
| Player | Product | Approach |
|---|---|---|
| AP2 / UCP | Cryptographic payment delegation; 60+ partners | |
| Visa | Intelligent Commerce | Tokenized credentials for AI; card-network constrained |
| Circle | Agent Wallets | Policy-controlled, permissionless USDC wallets |
| Stripe | MPP Session Model | Pre-authorized payment sessions with spending limits |
Layer 3: Discovery Underbuilt
An agent has a wallet. It has spending permission. What does it pay for?
This is the layer nobody is talking about — but it's where real economic activity starts. Before an agent can spend money, it needs to:
- Find services that solve its current task
- Evaluate quality, price, and reliability
- Try before committing budget
- Trust that the service will deliver
The settlement layer assumes agents already know what to pay for. In practice, discovery is the first step in every agent payment flow — and it's the layer with the fewest builders.
Discovery in Production: 324 Services, 319 Agents
On minia2a, the discovery layer is already live. The numbers tell the story:
Agents don't just need payment rails — they need a catalog. The top services by user count show what agents actually want:
| Service | Trial Users | What It Does |
|---|---|---|
| Gas Price | 158 | Real-time gas fees across chains |
| CAPTCHA Solve | 133 | Automated CAPTCHA breaking |
| Web Search | 84 | Agentic search with structured results |
| URL Finder | 80 | Discover endpoints and APIs |
| Polymarket Data | 53 | Prediction market prices and volume |
These aren't speculative use cases. Real agents are making real API calls for real tasks — gas lookups, CAPTCHA solving, web scraping, market data. The discovery layer is where they find these capabilities before any payment happens.
Why Discovery Matters More Than Settlement
Settlement is a commodity. There will be five viable settlement rails, and agents will route through whichever is cheapest or fastest for a given transaction. The x402 protocol already supports facilitator-agnostic routing — the agent picks the facilitator, not the service.
Discovery, on the other hand, has network effects:
- More services → more agents come to browse
- More agents → more developers list services
- More usage data → better quality signals (which services are reliable? which have the best uptime?)
- Better signals → agents make better spending decisions
This is the same dynamic that made app stores valuable. The OS (settlement) is important, but the store (discovery) is where the economic activity concentrates.
The Trial-First Pattern
One pattern that's emerging uniquely in the agent space: trial-before-payment as a discovery primitive.
Humans browse app stores by reading descriptions and reviews. Agents don't read — they test. A trial call ("give me one result, I'll pay if it's useful") is how agents evaluate services. This isn't a marketing gimmick; it's the only way an agent can assess whether an API endpoint solves its problem.
The stats bear this out: 11,683 free trials have been used across 319 unique agents — an average of 36 trials per agent. Agents are actively sampling the catalog, trying services before committing.
This trial-first pattern is something no settlement-layer product addresses. Coinbase CDP, Circle Agent Stack, and Stripe MPP all assume the agent already knows which service to call. The "how did the agent find this endpoint?" question is left unanswered.
What's Next
The stack will continue to fill in. Expect:
- Receipt standards — proving not just that payment happened, but what it paid for (on-chain tx is a settlement anchor, not a causality record)
- Quality signals — uptime, latency, and accuracy metrics that agents can query before spending
- Facilitator routing — agents picking the cheapest settlement path per transaction, agnostic to the discovery layer
- Budget-aware discovery — "find me a CAPTCHA solver under $0.005 with >99% uptime"
The settlement layer will commoditize. The authorization layer will standardize. The discovery layer — where agents find, try, and trust services — is where defensible network effects will form.
Data from minia2a production stats (August 9, 2026): 324 services, 319 unique agent users, 11,683 trials, 416K total requests. Settlement landscape data from public announcements by Coinbase, Circle, Stripe, Mastercard, AWS, and Solana Foundation.