# Agentic commerce: what it means for businesses that sell services

> Agentic commerce is the pattern where AI agents — acting on behalf of a human buyer — research, select, and complete purchases without step-by-step human approval. For businesses that sell services or data, this creates a new class of buyer: software that pays programmatically, expects machine-readable pricing and catalogs, and has no patience for checkout flows designed for humans. Businesses that expose agent-accessible APIs and accept stablecoin payments via protocols like x402 or MPP can capture this buyer class; those that don't will be bypassed.

6 min read · Updated 2026-06-09 · Topic: use-cases

Canonical: https://tempo.dataos.so/articles/agentic-commerce-for-businesses

A business that sells data, compute, software services, or API access has always had one type of buyer: a human, or a human-controlled organization. In 2026, a second buyer type is arriving at scale: AI agents acting autonomously on behalf of the humans who deployed them.

These agents do not browse, do not click, and do not create accounts. They call APIs, parse machine-readable responses, and pay programmatically. Businesses that are not set up to receive this type of buyer will not lose the sale to a competitor — they simply will not appear in the agent's consideration set.

## What agentic commerce actually looks like

Agentic commerce is not a single product category. It spans any transaction where an AI agent, rather than a human, makes and executes the buying decision. Current examples in production:

- An AI research agent calls a data enrichment API, pays per query in USDC, and incorporates the result into its analysis — without the human analyst ever seeing the API dashboard.
- A software deployment agent buys compute credits from a cloud provider's x402-enabled API, uses the compute, and the session closes when the job finishes.
- A customer support agent queries a product catalog API at $0.001 per request, paying from a funded wallet, and resolves customer questions at a cost per interaction lower than any human support tier.

The buyer in each case is software. The seller in each case is a business that exposed the right interface — a machine-readable catalog, a programmable payment endpoint, and a response structure the agent can use.

## The agent buyer's requirements

Selling to an agent buyer requires a different interface than selling to a human buyer.

| Human buyer | Agent buyer |
|---|---|
| Browses a website or catalog | Calls a discovery API or reads a machine-readable service description |
| Creates an account, stores payment credentials | Needs no account; payment is inline with the request |
| Pays by card, invoice, or bank transfer | Pays in stablecoins via x402 or MPP, or via Stripe's agent payment infrastructure |
| Waits for invoice and settlement | Expects payment to clear in under a second |
| Tolerates a checkout flow | Cannot navigate a checkout flow; abandons any step requiring human input |

The critical shift: the agent expects your service to declare its price in the HTTP response and accept payment in the same request cycle. If it cannot, it will skip your service and call one that can.

## Revenue models for agent-facing services

Three pricing patterns have emerged as the most practical for services selling to agents:

**Usage-based (per call / per token / per unit).** The agent pays for exactly what it consumes. This maps cleanly to x402 (one payment per request) and MPP Sessions (cumulative vouchers per session). A language model API charging $0.001 per 1,000 tokens, a data API charging $0.002 per query, or a compute provider charging by GPU-minute all fit this model.

**Outcome-based (per successful result).** The agent pays when it achieves a defined outcome. Intercom's $0.99 per resolved support ticket is a documented real-world example. This model requires the seller to define and verify the outcome before payment is released — more complex to implement but potentially higher-margin.

**Hybrid (base access fee + usage).** A small recurring access fee grants the agent a rate limit or priority tier; usage above that is billed per call. This provides the seller with baseline revenue predictability while still accommodating variable agent demand.

Flat subscriptions are the least natural fit for agent buyers because agent usage is task-driven and bursty. An agent that processes 50,000 API calls during a three-hour task and then goes dormant for days does not map well to a monthly seat license.

## Infrastructure for selling to agents: the Stripe approach

Stripe launched the Agentic Commerce Suite, announced in December 2025, to address the seller-side integration challenge. The suite includes:

- **Product catalog syndication** — merchants connect their catalog to Stripe, which distributes structured product and pricing data to agent networks via an ACP (Agent Commerce Protocol) endpoint.
- **Shared Payment Tokens (SPTs)** — a payment primitive scoped to a specific seller, bounded by time and amount. An agent carries an SPT as its payment credential; the seller verifies it without the agent transmitting full wallet credentials.
- **Checkout Sessions API** — handles tax and shipping calculations within the agent payment flow.
- **Stripe Radar integration** — distinguishes legitimate agent payments from fraudulent bot traffic using risk signal analysis.

Early adopters of the Agentic Commerce Suite include URBN brands, Etsy, Ashley Furniture, and Coach. The suite also integrates with Wix, WooCommerce, BigCommerce, Squarespace, and commercetools.

## The MPP Payments Directory

For businesses accepting stablecoin payments via MPP, the **MPP Payments Directory at mpp.dev** is the discovery layer. It listed over 100 services at Tempo's mainnet launch on March 18, 2026, with categories spanning model providers, developer infrastructure, compute platforms, and data services.

Being listed in the directory means an agent using MPP can discover your service programmatically as part of its task planning. Named participants at launch include Alchemy and Dune Analytics. Anthropic and OpenAI are also listed in the directory, per Tempo's mainnet documentation.

## Building an agent-accessible API

For a business that does not yet have an agent-facing interface, the integration path is:

1. **Define a machine-readable service description** — what the service does, what it costs, what it returns. This is the agent's version of your marketing page.
2. **Implement the HTTP 402 payment challenge** — using an x402 or MPP SDK, return a payment challenge when an unpaid request arrives.
3. **Accept stablecoin payments** — designate a wallet address (on Tempo for MPP, or on Base/Polygon/Arbitrum/Solana for x402) to receive payments.
4. **Return structured responses** — machine-parseable JSON, not HTML designed for human browsers.
5. **List in the relevant directory** — mpp.dev for MPP; x402.org for x402-compatible services.

The technical lift for steps 2 and 3 is modest for a service already built on HTTP. The SDKs for MPP (available in Rust, Python, and a CLI client) and x402 (TypeScript, Go, Python) abstract the payment verification logic.

## Market scale projections

McKinsey's QuantumBlack projected in October 2025 that agentic commerce will orchestrate $3 trillion to $5 trillion in global retail spend by 2030. Juniper Research's April 2026 forecast projects $8 billion in agentic spend in 2026 growing to $1.5 trillion globally by 2030. These figures are analyst projections with significant uncertainty, but they reflect the direction and order of magnitude of the opportunity being priced in across the infrastructure investment landscape.

For the payment protocol mechanics that underpin this commerce pattern, see [The Machine Payments Protocol (MPP), explained](/articles/machine-payments-protocol-mpp) and [How AI agents use stablecoins to pay for APIs](/articles/ai-agents-pay-apis-stablecoins). For what stablecoin acceptance means for retail e-commerce specifically, see [Stablecoin payments for e-commerce](/articles/stablecoin-ecommerce-payments).

## FAQ

**What is agentic commerce?**

Agentic commerce is the buying and selling of goods or services where an AI agent acts as the buyer — autonomously researching options, making decisions, and completing payment — on behalf of a human. The human sets a goal and budget; the agent handles the rest.

**How is selling to an AI agent different from selling to a human?**

An AI agent cannot navigate a checkout flow, fill out a form, or create an account. It interacts with your service through APIs and expects machine-readable pricing and catalogs. Payment must be programmable — stablecoins via x402 or MPP — rather than a card with a human present.

**What is a Shared Payment Token (SPT)?**

Shared Payment Tokens are a payment primitive introduced by Stripe's Agentic Commerce Suite. An SPT is scoped to a specific seller, bounded by time and amount, and observable throughout its lifecycle — allowing a seller to accept payment from an agent without exposing full payment credentials.

**Do I need to accept stablecoins to participate in agentic commerce?**

Not necessarily — Stripe's Agentic Commerce Suite works with existing card and wallet infrastructure. But agents that use x402 or MPP pay in stablecoins, and those protocols give the agent a standard way to discover and pay for services. Accepting stablecoins via these protocols makes your service discoverable to that buyer pool.

**What revenue models work for selling to agents?**

Usage-based pricing (per API call, per token, per query) maps most directly to how agents consume services. Outcome-based pricing (per successful action or result) is emerging. Flat subscriptions are less natural for agents because agent usage patterns are bursty and task-driven, not steady-state.

## Sources

1. [Stripe — Introducing the Agentic Commerce Suite](https://stripe.com/blog/agentic-commerce-suite)
2. [Salesforce — What is agentic commerce?](https://www.salesforce.com/commerce/ai/agentic-commerce/)
3. [Tempo — MPP Payments Directory (mpp.dev)](https://tempo.xyz/blog/mainnet)
4. [Tempo — MPP Sessions](https://tempo.xyz/blog/mpp-sessions)
5. [x402 protocol documentation](https://docs.cdp.coinbase.com/x402/welcome)

---
Neutral, sourced explainer from tempowiki. Index: https://tempo.dataos.so/llms.txt