# agents.u.cash — Use Cases

What AI agents actually sell, buy, and operate through U.CASH Pay once payments are HTTP-native and rail-agnostic. This catalog maps each use case to the buyer, the pricing model it enables, and where it lands on the roadmap. **Every flow here is non-custodial** — funds stay in agents' own wallets; the platform verifies, relays, and indexes, never holds (see [ARCHITECTURE.md](ARCHITECTURE.md)).

## Use cases at a glance

| Category | Example | Buyer | Pricing 402 uniquely enables | Phase |
|---|---|---|---|---|
| Pay-per-call API / data | $0.002/query research API | agent / human | per-request, sub-cent | v1 |
| Pay-per-inference | $0.0001/token LLM | agent | streamed/continuous | v1 |
| Compute / GPU (TEE-attested) | $0.06/min | agent | per-second, attested | v2 |
| **Agent skills / MCP servers** | a monetized MCP tool | agent | per-call | v1 |
| Curated datasets / feeds | market data | agent / human | subscription + per-pull | v2 |
| Verification / oracle | "is this URL safe?" | agent | per-attestation | v2 |
| Search / retrieval / RAG | $0.005/search | agent | per-query | v1 |
| **Escrow / arbitration agents** (smart-contract) | release-on-delivery | agents | per-deal percentage | v2 |
| Monitoring / alerting | uptime with paid escalation | business | per-event | v2 |
| **Routing / orchestration** | an agent-of-agents | agents | spread / commission | v2 |
| Agent-operated storefront | goods or services | humans (card) + agents (402) | mixed | v2 |
| **Agents paying humans** | tasks, data-labeling, "captcha-for-agents" | agent → human | micro-payouts cards can't do | v2 |

## Agent-as-seller: what agents sell

Eleven concrete categories:

1. **API / data access** — the canonical 402 case. A vision API charges $0.005 per image classification instead of a flat enterprise fee. Cards cannot do $0.005 (a ~$0.30 fixed floor consumes 98% of it).
2. **Pay-per-inference / per-token** — replace subscription tiers with metered billing. Streamed payment (per token, per GPU-second) is the killer pattern subscriptions can't express.
3. **Compute / GPU / TEE** — attested per-second billing. Trust comes from remote attestation bound to the payment.
4. **Curated datasets and feeds** — subscription plus per-pull, with royalty splits flowing to upstream contributors.
5. **Agent skills / MCP servers** — *the* canonical 402 case as the agent ecosystem adopts the Model Context Protocol. Highest near-term demand: any tool an agent calls can be priced per-call.
6. **Generated content, code, art** — pay-per-artifact.
7. **Verification / oracle services** — paid attestations that other agents consume as a trust input.
8. **Search / retrieval / RAG** — per-query.
9. **Escrow / arbitration agents** — a *new agent class*: a neutral third party (or smart contract) paid per deal. Funds sit in a **smart contract** that neither counterparty — nor the platform — controls, released on a verifiable condition. The trust graph becomes recursive — every layer is metered, and the platform never holds escrow.
10. **Monitoring / alerting** — paid escalation tiers.
11. **Premium routing / orchestration** — the "agent-of-agents" that composes specialists and takes a spread. The highest-leverage *emergent* use case. Don't build it — *enable* it.

## Agent stores

An **agent store** is a catalog of priced resources. It maps cleanly onto existing platform primitives: one agent-app is one store; a priced resource is one checkout; fulfillment is the existing webhook system (already event-driven). One store has two faces:

- **Human face:** a branded landing and pricing page. The human pays by **card or by 402**.
- **Agent face:** `402 → pay → retry with the payment header → JSON`. Plus machine-readable discovery per merchant — **`/catalog.json`** (schema.org Product JSON-LD), **`/catalog.txt`**, and **`/.well-known/ucp`** (the UCP profile) — so an agent can discover and consume with zero human-readable docs.

A useful side-effect: the platform's existing spend-tier unlocks double as **Sybil resistance** — "$0.01 to unlock 100 free calls" raises the cost of farming free tiers to unprofitable.

## Ecommerce integration

Four modes, in increasing ambition:

- **Agent-operated storefront** — an agent runs a shop humans buy from by card, supplied by other agents over 402.
- **Autonomous inventory procurement** — *the killer pattern.* An agent **buys wholesale via 402 (USDC) and sells retail via Stripe (card)**, earning the spread in its own wallet — both sides settle peer-to-peer and the platform holds nothing. Cards literally cannot serve as the cost side for an autonomous buyer.
- **Agent-run dropshipping** — agents orchestrate suppliers and fulfillment end to end.
- **Affiliate agents paid per conversion in real time** — $0.40 micropayouts that card rails cannot move economically.

Existing leverage: U.CASH Pay already integrates billing systems like WHMCS and has the hooks for Shopify/WooCommerce-style flows. Reuse is roughly 80%; the new surface is the UCP catalog per merchant, a 402 endpoint per resource, and a browse UI.

## Patterns uniquely enabled by 402 / micropayments

Sub-cent pricing · per-request · **streamed/continuous** (tokens, GPU-seconds) · usage-based without a card on file · **pay-per-outcome** (escrow plus a verifier attestation) · bid/ask double auctions · **nested sub-cent royalty splits** · cross-border at parity (~$0.001 gas versus 3.5%+ international card plus FX) · reversible-by-design escrow.

## End-to-end vignettes

**A. Research agent, pay-per-call.** An agent indexes a domain and sells queries at $0.002 each. A buyer agent hits the endpoint, receives a 402 with a `detect` entry (the seller's wallet + an exact amount), pays that amount on-chain direct to the seller's wallet, and gets JSON. The platform detects the payment on-chain, confirms, and settles — funds moved **direct buyer→seller**; the platform debits its UCASH-rate fee from the seller's prepaid credit, indexes the result, and fires the seller's webhook. The platform never holds the payment.

**B. Inference agent, streamed per-token.** An LLM agent charges $0.0001/token. A long generation accrues; the buyer settles per call (or small batches) by paying `detect` entries direct to the seller's wallet — **each payment moves funds direct buyer→seller**, never pooled, so per-token cost stays sub-millicent.

**C. GPU agent, attested per-second.** A compute agent bills $0.06/min inside a TEE, with the attestation bound to the payment so the buyer can verify the work before release.

**D. Agent-operated store, two-rail.** An agent buys inventory wholesale from supplier agents over 402 (USDC) and resells to humans via its own Stripe Checkout, earning the spread in its own wallet — each side settles peer-to-peer, nothing held by the platform.

**E. Orchestrator agent, spread.** A routing agent composes four specialists into a pipeline, each hop a paid 402 call. It takes a 12% spread over its subcontractors — every participant settles **peer-to-peer on-chain**, with the splits enforced atomically by Builder Code at settlement (not reconciled later by a custodian).

## What agents can buy (the other side)

- **From other agents** — any service above, paid autonomously in USDC over 402. Multi-agent pipelines, procurement, escrow.
- **From humans** (the un-thought-of market) — micro-tasks, data labeling, human verification. The agent pays from its **own wallet** (USDC on-chain, or via the human's own PayPal/processor); the platform only relays the destination and never holds funds — a non-custodial two-sided micro-labor market cards cannot match.

## What is missing today

Already shipped since this was written: the 402 middleware and the non-custodial facilitator (exact-verify / exact-settle, gasless by default) and the machine-readable UCP catalog. Still open: usage metering; agent receipts/invoices; crypto recurring/streamed payments; a smart-contract escrow primitive; and agent identity/attestation. Most of the foundation — merchant accounts, multi-store, webhooks, card plus 30+ network crypto, and non-custodial payment detection — is already there.

## Read next

- [ARCHITECTURE.md](ARCHITECTURE.md) — how a 402 challenge carries every rail, and how settlement is verified per rail.
- [VISION.md](VISION.md) — the thesis and the agent economy at scale.
- [ROADMAP.md](ROADMAP.md) — which of these land in v1, v2, v3.