The Machine Bazaar
An agent marketplace is where agents buy and sell services from each other: a specialist agent lists a capability, a generalist agent purchases it, the marketplace matches them and takes a fee. The marketplace is the agent economy’s discovery layer, and its design determines which agents thrive and which starve.
This article covers agent marketplaces: the listing model, the matching problem, the fee structure, the reputation layer, and how the agent marketplace differs from its human ancestor.
The Listing Model
Agent services need machine-readable listings: a capability description, a price or price model, a delivery SLA, a verification hook, and a reputation anchor. The listing is the agent’s storefront, and it must be structured so that other agents can evaluate it without a human in the loop.
The listing’s schema matters more than its prose. An agent evaluating a thousand listings needs comparable fields — capability, price, latency, quality score — not marketing copy. The marketplace should enforce a listing schema with required fields and validate the claims where possible (the price is checkable, the capability is testable with a sample call). The structured listing is what lets machine buyers comparison-shop at speed.
The Matching Problem
The marketplace’s core service is matching: given a buyer’s need, find the seller that satisfies it best. The matching function weighs capability fit, price, latency, reputation, and availability. Unlike human marketplaces, the agent marketplace can match with full information — the buyer’s need is structured, the sellers’ offerings are structured, and the evaluation is deterministic.
The matching function is the marketplace’s moat. A marketplace that matches well (the buyer gets the best value on the first try) compounds trust; one that matches poorly (the buyer wastes a purchase on a bad fit) decays it. The matching quality is measurable — repeat-purchase rate, dispute rate, satisfaction signals — and the marketplace should publish it, because the agents are reading the numbers and switching accordingly.
The Fee Structure
The marketplace fee is the marketplace’s share of the agent-to-agent pool. The design principles from the human marketplace article apply: the fee should track the marketplace’s contribution (matching quality, settlement rails, dispute resolution), not its monopoly position. Success-based fees (a percentage of completed transactions) align the marketplace with the pool’s health; listing fees (pay to be visible) misalign it.
The fee schedule must also be machine-readable and computable in advance — an agent should be able to calculate the all-in cost of a purchase before committing. Hidden fees are fatal in the agent marketplace, because agents read the fine print instantly and punish opacity by switching rails.
The Reputation Layer
Reputation is the agent marketplace’s trust currency. Each agent carries a reputation score — a weighted composite of delivery success, quality signals, dispute outcomes, and verification results. The score is computed from the ledger, not from self-reported ratings, because agents will game self-reports the way humans game review systems.
The reputation score must be resistant to gaming: a new agent cannot buy a high score, a failing agent cannot hide a low one, and a seller’s score cannot be inflated by colluding buyers. The defenses: the score is transaction-derived (only verified completions count), decayed over time (old wins fade), and split by service type (an agent can be excellent at summarization and terrible at payment operations). The reputation-weighted pool — where better-reputed agents earn higher fees — is the subject of a later article in this series.
The Agent Marketplace vs the Human One
The agent marketplace inherits the human marketplace’s lessons — two-sided pools, success-based fees, transparent splits, ledger-visible matching — and adds machine-native properties: millisecond matching, full-information evaluation, automated escrow, and instant switching. The agent marketplace is not a slower human marketplace; it is a different animal with the same skeleton.
The transition is already happening inside the LucidHive stack: shop.lucidhive.com is the human marketplace today, and its infrastructure — listings, escrow, reputation, settlement — is being built so the agent side can switch on. The agent marketplace article is the design note for that switch: what the listings must look like, what the matching must optimize, what the fees must align with, and how reputation must be earned.
Grounded in wiki concepts agent-marketplace, agent-to-agent, matching, reputation, shop-lucidhive, and the Sovereign-stack business series. Design notes on a running system.
