Consumer shopping agents, agent-to-agent marketplaces, NFT and token commerce, B2B procurement, and provenance-verified commerce. The smallest anchored category in the Phase 2.2 sample, but situated within one of the fastest-expanding surfaces of the agent economy.
A Commerce Agent is an autonomous software agent whose primary function is to participate in marketplaces — discovering products or services, comparing options, negotiating, completing purchases, and managing the end-to-end commerce lifecycle. Distinct from Financial Agents (which transact on financial instruments themselves) and Orchestration (which coordinates other agents), Commerce Agents are economic actors transacting against catalogues of goods and services.
Commerce is the smallest of the six anchored categories in the Phase 2.2 sample, with 20 classified agents (2.6%). The size in the sample is misleading about the scale of the surrounding category: agentic commerce is one of the fastest-growing areas of the broader AI economy, with mainstream production deployments live across ChatGPT Instant Checkout, Coinbase Agent.market, Mastercard Agent Pay, Visa Intelligent Commerce, and Amazon Buy for Me. McKinsey projects the channel could reach $1 trillion in US retail revenue by 2030.
In the Phase 2.2 classification run, Commerce Agents were anchored on Seaport and SeaDrop — OpenSea's NFT marketplace infrastructure. An agent qualified as Commerce if its on-chain history showed direct interaction with these contracts. The methodology is documented in the POC summary; subsequent classification phases are expected to expand the anchor set substantially.
The POC anchor scope (Seaport, SeaDrop) captures one specific commerce pattern — NFT marketplace participation on Base — and produces a small but well-defined cohort. The wider Commerce category is much larger and increasingly multi-rail: agent-driven retail purchases settling on card networks (ChatGPT Instant Checkout via Stripe ACP, Mastercard Agent Pay), agent-to-agent marketplaces settling in stablecoins (Coinbase Agent.market via x402), B2B procurement on traditional rails, and agentic checkout integrations across Shopify, Amazon, and Visa TAP. Future classification phases will need to bring more of this surface into the anchor set; the current 20-agent cohort should be read as a faithful narrow slice rather than a comprehensive census of agentic commerce.
Agents acting as proxies for human consumers, executing the discovery-to-purchase lifecycle from a goal rather than a click. ChatGPT Instant Checkout, Amazon Buy for Me, Perplexity Shopping are the production deployments.
Venues where agents transact with each other rather than with humans. Coinbase Agent.market on x402 is the headline production example; Anthropic's Project Deal pilot is the most-discussed research deployment.
Agents trading on Seaport / SeaDrop-style NFT marketplace infrastructure. The POC anchor pattern, and the cohort the 20 classified agents fall into.
Autonomous procurement, RFQ handling, and contract negotiation. Forrester predicts 20% of B2B sellers will face AI-led quote negotiations in 2026, compressing negotiation cycles from days to minutes.
Agents transacting on branded or regulated commerce with verifiable credentials anchoring product authenticity, authorised reseller status, or compliance. MolTrust's MT Salesguard pattern is the worked example.
The platforms below are well-known illustrations of the Commerce category as defined above. OpenSea Seaport is both the Phase 2.2 POC anchor and the canonical NFT marketplace pattern; the others span the wider agentic commerce landscape that future classification phases are expected to capture. Several entries are protocols and platforms rather than discrete agents — the unit of analysis in Commerce is increasingly the rail, not the registrant.
The Phase 2.2 anchor for the entire Commerce category. Seaport is OpenSea's NFT marketplace protocol; SeaDrop is the launch-and-claim primitive layered on top. The 20 Commerce agents in the AHM sample qualified through direct interaction with these contracts. Representative of the longer-running on-chain commerce pattern that pre-dates the current agentic-commerce wave: agents bidding, listing, sweeping, and minting NFTs against an established marketplace primitive.
Launched April 2026 as an app store for autonomous agents that pay each other in stablecoins via x402. By late April 2026, the broader x402 ecosystem reported approximately 69,000 active agents, 165 million transactions, and roughly $50 million in cumulative volume — average value per call calibrated for sub-cent micropayments. Representative of the genuinely agent-native commerce pattern: agents discovering, negotiating, and paying each other without humans in the per-transaction loop.
Production deployment of OpenAI's Agentic Commerce Protocol, co-developed with Stripe. Live since September 2025; over 1 million Shopify merchants opted in by early 2026. Issues Shared Payment Tokens (SPTs) bound to specific merchant and dollar amount, time-bounded and single-use. Note: this is OpenAI's "Agentic Commerce Protocol" — distinct from Virtuals' "Agent Commerce Protocol" which anchors the Orchestration category. The naming collision is unfortunate but each is a different protocol in a different rail.
Co-developed by Google, Shopify, Etsy, Walmart, Target and Wayfair, with 20+ endorsers across payments and retail. Launched January 2026. Standardises how agents discover what merchants can do via /.well-known/ucp manifests, and how transactions flow when an agent can act fully autonomously versus when it must hand off to a human via the Embedded Checkout Protocol. The "HTTP of commerce" framing — UCP for context, x402 for settlement when stablecoin rails are used.
Research pilot announced April 2026. Anthropic created a classified marketplace where AI agents represented both buyers and sellers, striking real deals for real goods and real money — 69 employees with $100 budgets in gift cards, 186 deals made, $4,000+ in value. A self-selected pilot rather than production infrastructure, but a worked demonstration that agent-on-agent commerce can produce useful economic activity at small scale. Included as the experimental edge of the category.
Goal-based purchasing instead of session-based browsing. The headline shift is from "I'm buying this product" to "solve this problem within my budget." A consumer describes intent — "trail-running shoes under $150 that arrive Friday" — and an agent handles discovery, comparison, and checkout. The change in unit of work is what makes agentic commerce structurally different from AI-assisted shopping: the agent owns the transaction, not just the recommendation.
Machine-to-machine commerce at sub-cent granularity. Where consumer commerce settles on card rails for $50–$500 transactions, agent-to-agent commerce settles on stablecoins for sub-cent transactions. x402's reported $600 million annualized volume with average value per call of about 30 cents reflects this micropayment-native pattern. APIs, compute resources, data feeds, and other digital services become acquireable at the granularity of a single function call.
Compressed B2B negotiation cycles. Procurement and quote negotiation that used to take days reduces to minutes when both sides are software. The Forrester prediction that 20% of B2B sellers will face AI-led negotiations in 2026 reflects this — and the buyers gain by parallelising negotiation across many suppliers simultaneously, which a human team could not do at the same scale.
Provenance-verified commerce for regulated and branded contexts. When a shopping agent buys on behalf of a consumer, the consumer cannot easily verify counterparty, product authenticity, or compliance status. Verifiable credential infrastructure (W3C VCs, EAS attestations) layered into commerce protocols lets agents check brand authorisation, product provenance, and regulatory status before committing — closing a trust gap that pure speed-of-execution doesn't address.
Composite 0–100 health score blending behavioural consistency, counterparty integrity, and operational stability. Calibrated for commerce flows where the unit of trust is often the merchant or marketplace, not the individual agent.
Operational health scan: failed-transaction patterns, gas-spend anomalies, post-purchase failure indicators, dispute and refund history where data is available.
Surface merchant relationships, credential-redemption patterns, and cross-jurisdiction transaction graphs. Designed to expose merchant impersonation clusters and cross-border compliance anomalies.
Smart money labels via Nansen, PnL enrichment, and consistency checks between observed pricing patterns and counterparty behaviour. Useful for shopping agents and treasury-level commerce both.
@misc{ahm_taxonomy_commerce_2026,
title = {Commerce Agents --- AHM Taxonomy v1},
author = {{Agent Health Monitor}},
year = {2026},
month = {May},
url = {https://intelligence.agenthealthmonitor.xyz/taxonomy/commerce},
note = {Accessed: 2026-05-07}
}
Classification methodology and category boundaries are documented in the AHM Taxonomy v1 POC summary.
Counts reflect the Phase 2.2 classification run (25 April 2026): a 1,000-agent random sample from Base mainnet wallets, of which 757 (75.7%) were classified across 6 anchored categories. Updated as new classification phases complete.