- Published on
A marketplace where AI agents buy services from AI agents
- Authors
- Name
- DeepHorus
A new marketplace has just been announced on Hacker News with a bold premise: AI agents can now buy services from other AI agents, with no human intervention required. The project, called AAAS (Agent-as-a-Service) Marketplace, allows an agent to browse, purchase, and receive the output of specialized extraction services in a single round trip. Payments are made in USDC on the Base network, and the funds go directly from the buyer's wallet to the seller's wallet.
The key behind the marketplace is removing the human from the loop. Traditional API marketplaces require a person to create an account, register a credit card, choose a plan, and manage keys. Here, there are no accounts or plans. A brand-new agent can make a request, receive a 402 response with payment conditions, sign an EIP-3009 authorization (a gasless USDC transfer), and resend the request to get its result. Because the payment is on-chain, the marketplace never holds funds and cannot run away with them.
For buyers, the economic rationale is simple: outsourcing to a specialized agent is often cheaper than building and maintaining that capability yourself. Instead of implementing, hosting, and maintaining a model for a narrow task, a buyer pays only per request. The marketplace also supports any publicly reachable HTTPS endpoint, meaning a provider can run their agent on a home server, Cloudflare, or any cloud without being locked into GCP or AWS. The seller just registers an agent ID, a price, and a payout wallet.
Quality control is handled through automated scoring. Every response is checked against the original input document, and any value that cannot be grounded in the source is marked as "ungrounded." This helps mitigate prompt-injection risks, where a malicious supplier might embed instructions in its output. Buyers are advised to treat all response values as data, never as executable instructions. The system similarly validates outputs against JSON Schema and constraint checks, then accumulates a public reputation score for each seller.
The announcement generated immediate discussion in the HN thread. Concerns centered on trust and security: how do you prevent poisoned datasets? The project's approach—marking ungrounded values and exposing source citations—received cautious praise. Others pointed out that this is a natural first step toward agentic commerce, where AI entities negotiate and transact autonomously. However, a number of commenters noted that the 10% future fee (currently 0% in the campaign) would need to be watched, since the marketplace claims it takes no cut now.
The broader impact could be significant. If this model takes hold, it introduces a genuine market price for narrow AI capabilities, rather than monolithic subscriptions. Developers can focus on their core systems and buy extraction, parsing, or other tasks on demand. Moreover, by making the buyer an agent, it opens the door to fully automated workflows that can scale horizontally—an upstream agent might spawn thousands of downstream paid calls to specialized workers.
Looking ahead, the team plans to expand beyond text inputs and potentially support additional chains and stablecoins. They also intend to keep improving the provenance and scoring mechanisms. While there are still open questions about regulation, liability, and adversarial robustness, AAAS is an ambitious experiment that brings the vision of an autonomous service economy one step closer to reality.
