The Discovery Layer Is the Open Battleground in the Agent Economy

Three things happened in the last 72 hours that clarify the agent payment landscape better than any analyst report could:

1. Stripe is buying OpenRouter for $10 billion — building the Visa of the AI era by owning the "meter → route → settle" stack.
2. AIsa raised $6.5M from Alibaba and Tribe Capital — 50,000 agents onboarded, #1 x402 seller, processing millions of microtransactions.
3. Coinbase's x402 registry quietly hit 14,865 services — across 1,561 hosts, auto-indexed, zero curation.

These three data points reveal the same thing: the payment rails are being commoditized, and the discovery layer is the open battleground.

The Stack Is Layering, Not Consolidating

By August 2026, the agent payment infrastructure has settled into five distinct layers:

Stripe's OpenRouter deal essentially locks up Layer 3 (Routing). The protocols (Layer 1) and settlement chains (Layer 2) are already mature with 169M+ x402 payments processed. Banking (Layer 4) is being built by well-funded startups.

But Layer 5 — Discovery — has no dominant player. And that's the layer that decides which APIs agents actually call.

The Discovery Layer: Four Approaches, No Winner

Right now, there are four competing models for agent API discovery:

1. Auto-Indexing (Coinbase Bazaar)

Coinbase's x402 registry auto-discovers services by crawling the network. 14,865 services across 1,561 hosts. The upside: exhaustive coverage. The downside: zero curation. No ratings, no health checks, no quality signals. An agent searching for "token security" gets dozens of results with no way to distinguish a reliable endpoint from one that's been down for weeks.

Model: Google-style crawl-and-index. Problem: ranking by raw request volume is gameable. The top results aren't necessarily the best — they're the ones with the most bot traffic.

2. Protocol-Specific Directories (mpp.dev, L402 directories)

Tempo's mpp.dev covers MPP services exclusively — high quality, institutional listings including Anthropic, OpenAI, and Dune Analytics. Lightning-focused directories like Satring cover L402. Each is excellent within its protocol but invisible to agents using other payment standards.

Model: Walled garden. Problem: the agent economy won't be single-protocol. Agents will use x402 for some calls, MPP for others, L402 for Lightning-native services.

3. Multi-Protocol Intelligence (Cinderwright, Satring)

Cinderwright indexes across x402 (1,457 services), MPP (92), and L402. Offers paid market intelligence endpoints — gap analysis, category deep dives, competitive reports. Satring started as an L402 directory (562 services) and now accepts x402 payments too. Both are moving toward cross-protocol coverage.

Model: Data platform. Problem: indexing services is different from making them discoverable and usable by agents.

4. Curated + Trial-First (minia2a)

300 curated services, every single one with a free trial (15 calls). Agents don't just see a list — they test the endpoint before committing capital. The catalog is smaller than Coinbase Bazaar's 14K but every listing is verified: uptime monitored, schema documented, pricing transparent. EU-hosted (AWS Ireland) for GDPR compliance.

Model: App Store. Problem: curation doesn't scale as fast as auto-indexing. The catalog needs to grow while maintaining quality.

Why Discovery Wins When Settlement Gets Commoditized

This isn't theoretical — we've seen this movie before:

2009–2012: Payment rails were the bottleneck. Stripe won by making payments easy.
2026: Payment rails are solved. x402, MPP, USDC on Base — the rails work. The new bottleneck is discovery.

When every agent can pay for any API with a single HTTP 402 header, the question stops being "can I pay for this?" and becomes "which of these 14,865 APIs should I pay for?"

The data from minia2a's own platform supports this. Across 6,911 trials from 316 unique agent users:

The pattern is clear: utility endpoints acquire users (captcha: 133 users), data endpoints retain them (recall: 14.5 calls/user, fear-greed: 23.0 calls/user). A discovery layer that only lists APIs without understanding this usage pattern can't guide agents to the right tools.

The Three Characteristics of a Winning Discovery Layer

Based on what's working and what's not across all four models:

1. Trial-First, Not List-First

Agents don't browse. They query. A discovery layer that forces agents to read documentation before trying an endpoint loses to one that lets them call the endpoint immediately with a free trial. The decision to pay happens after the first successful call, not before.

2. Quality Signals, Not Just Volume

Coinbase Bazaar's ranking by raw request count is trivially gameable — a service that gets botted looks more popular than a genuinely useful one. Discovery needs multi-dimensional signals: uptime, latency, trial-to-paid conversion, user retention, schema completeness.

3. Cross-Protocol, Not Walled Garden

The agent economy won't settle on a single payment protocol. x402 dominates today (169M transactions), MPP is growing (Stripe-backed, institutional quality), and L402 has a dedicated Lightning-native community. A discovery layer that only works with one protocol is building on rented land.

What This Means for Agent Developers

If you're building an agent that calls paid APIs:

1. Don't couple to a single discovery source. Use agents.json for structured discovery, but also check cross-protocol indexes. The landscape will consolidate, but it hasn't yet.
2. Prefer trial-first platforms. Every API call costs money. Platforms that give you free trials before you commit let you validate utility without burning budget.
3. Watch the curation layer. Auto-indexed directories (Coinbase Bazaar) will grow faster. Curated directories (minia2a, Satring) will be more reliable. Your agent needs both — breadth for discovery, depth for trust.
4. Build for multi-protocol from day one. Your agent should be able to pay via x402 or MPP or L402 depending on which API it's calling. Payment abstraction is the next layer to build.

The Next 12 Months

Here's my prediction for how the discovery layer evolves:

• Q3 2026: Coinbase Bazaar adds basic curation signals (uptime, ratings). The auto-index approach gets a quality layer.
• Q4 2026: First acquisition in the discovery space — likely a curated directory bought by a settlement provider (Circle? Stripe?) looking to own the full stack.
• H1 2027: Emergence of "agent SEO" — API providers optimizing their listings for agent discovery algorithms, just like websites optimized for Google in 2005.
• H2 2027: The discovery layer consolidates to 2-3 players. The winners will be the ones that combine breadth (indexing everything) with depth (curation, trials, quality signals).

The payment rails are built. The routing layer has its $10B champion. The banking layer is well-funded. The only question left is: who will agents ask when they need to find an API?

Data: minia2a /api/stats, August 4, 2026. 300 services, 6,911 trials, 316 agent users, 342,692 total requests.