Uncovering the Truth About Agent Commerce, Payments, and Infrastructure
Decoding Agent Commerce, Payments, and Infrastructure: The Reality
Over the past year, I've been building infrastructure for the Agent economy, engaging with major players like Stripe, Visa, Coinbase, Google, and dozens of startups. A clear conclusion emerges: true, large-scale demand does not yet exist. Startups face structural challenges.
Data points illustrate this gap. Stripe's Agent commerce platform has over 1,000 merchants but only single-digit transacting agents. Visa's Agent payment token requires 9-month KYC and a $250M revenue threshold, accessible only to giants like Amazon. On-chain analysis reveals actual daily Agent transaction volume is around $17k, half of which are test transactions.
The article analyzes four potential markets:
**1. Agent-to-Merchant (A2M):** Current AI shopping UX is often inferior to traditional e-commerce for visual, comparison-heavy purchases (clothing, electronics). Chat interfaces are a step back. Real merchant interest is defensive "Agent Engine Optimization," fearing future obsolescence, not current demand. Potential exists in high-frequency, low-decision purchases (e.g., food delivery) or simplifying terrible UX (complex checkouts, non-native shoppers), but these require massive consumer distribution channels dominated by giants like DoorDash and Amazon.
**2. Agent-to-API (A2A):** Developers already have subscriptions and billing for core APIs (compute, data). The argument for micro-payments via crypto for sub-dollar API calls is addressed by pre-paid balances today. The deeper issue is supplier resistance; major SaaS firms rely on enterprise contracts, not fractional cent pricing. Opportunity lies in the long tail of niche services, but this is a smaller market catering to developers, a historically low-paying group.
**3. Agent-to-Agent (A2A):** This remains a theoretical long-term vision with near-zero current transaction volume. It involves unique challenges: discovery, trust, negotiation, dispute resolution. When it materializes, it will require a fundamentally new settlement infrastructure for high-speed, variable-value, multi-party transactions. It's a real long-term bet, but not the current market.
**4. Agent-to-Finance (A2F):** This is the only category with existing, paying demand. Integrating AI into financial workflows (trading, portfolio management) is a natural evolution and enables new capabilities like autonomous rebalancing. However, competition favors incumbents with regulatory licenses, compliance infrastructure, and existing client relationships.
**The Real Issue:** Why is infrastructure still being built? Incumbents can afford long-term bets, and payment companies see every problem as a nail for their payment hammer. However, payment is just one piece. The core challenge is *coordination*—orchestrating work between Agents and humans, verifying outcomes, and settling results. Payment is part of settlement, which is part of coordination. Companies that solve the coordination problem will subsume payments, not the other way around.
Startups lack the infinite runway of giants and must find today's real market, which, after a year of exploration, lies outside these four categories—in an area with real, growing, and underserved activity.
marsbit2 h fa
