Skip to content

561 Group · Research

Agent-to-Agent Commerce: When Software Buys From Software

Agent-to-agent commerce is exchange in which an autonomous software agent initiates and settles a purchase, usually per call, with no human approving each.

Research

Agent-to-Agent Commerce: When Software Buys From Software

Agent-to-agent commerce is economic exchange in which an autonomous software agent, not a human at a checkout, initiates and settles a transaction with another agent or service, typically a per-call payment for a machine-payable API. The buying decision, the payment and often the consumption of the result all happen inside a software process, with a human present only as the operator who deployed the agent and funded its wallet.

How does it work?

A seller builds a resource priced for per-call purchase, for example behind an HTTP 402 challenge as defined by x402. An agent working on a task finds the resource through its own tool search, an MCP server list or an x402 directory such as the x402 Bazaar, meets the challenge, pays from its own wallet through a facilitator, and continues, the way a conventional program calls a library function. Cheap micropayment settlement is the enabling condition.

Why does it matter?

It is the buy-side complement of the machine-payable API: the API is the shop, agent-to-agent commerce is the shopper that arrives without a person. If it grows, it changes who is a customer: sellers could serve strangers they never meet, billed per request, with no account relationship. Large payment, cloud and card firms joined the x402 Foundation in 2026, and several rival agent-payment protocols exist, though they address different questions: whether a person authorized a checkout, how a request carries its payment, and whether an agent is who it claims to be.

How big is it really?

Smaller and thinner than the promotional framing of an "AI agent economy" suggests. Chainalysis' account of x402 on Base found that a large share of headline transaction growth traced to speculative pay-to-mint activity around a token (the PING token, 2025-10-14 to 2025-10-20), not agents autonomously buying services. Filtered estimates place plausible x402 commerce between about $15 million and $26 million over sixteen months, and a TRM Labs analysis bounds the share plausibly from AI agents at 0.6% to 7.5% of filtered value, because agency is not visible on chain. A script and an agent leave identical records. See the x402 economy for dated figures and x402 Bazaar and agent-market resource discovery for what catalogs can and cannot show.

What are its limits?

  • Discovery is the hard part. A resource is worthless to this buy side unless an agent's tool search or directory query surfaces it. Our own paid x402 seller was found by at least two strangers, sold a few times, then went quiet for a month, with no mechanism in the record explaining why.
  • Demand skews to utilities. The most-called catalog resources on 2026-08-27 were thin blockchain data lookups, each with roughly one call per payer, which suggests sampling more than integration.
  • Product fit matters more than the rail. Free overlapping tools and generous subscription free tiers beat a $0.04 price for our seller's niche product.
  • Legal and security questions remain open: money-transmitter status, micro-income tax, no-chargeback disputes, and published attacks on payment-to-service binding.

The 561 Group's own market surface is market.561.group.

Contents

Explore

Tagged

  • agent-commerce
  • x402
  • machine-payable-api
  • two-sided-market
  • measurement