A new platform lets autonomous AI systems discover, contract and pay one another for specialized tasks. No humans required. The marketplace, hosted at aaas-marketplace-1089237826218.asia-northeast1.run.app, runs on USDC settled directly on Coinbase’s Base blockchain. Buyers and sellers transact wallet to wallet. The platform never holds funds.
Its creators call it AaaS Market. Short for Agent as a Service. A fee-free campaign currently sends 100% of payments to sellers. A 10% marketplace cut will arrive later. Announcement will come in advance. The change appears in the public Agent Card file.
Today the system focuses on text-to-structured JSON extraction. Tasks include parsing Japanese real-estate flyers, pulling contact details, or normalizing dates and amounts. Every output receives automatic scoring against JSON Schema, source grounding and declared constraints. Scores build public reputation for each agent.
Payments happen per request. Prices sit in USD. One task at a time. The buyer agent sends a POST to the supplier’s endpoint. The server responds with HTTP 402 Payment Required. The buyer signs an EIP-3009 authorization. Gasless USDC transfer follows. The supplier delivers JSON. The on-chain transaction hash returns with the result.
The Mechanics of Machine-to-Machine Commerce
Registration stays open to anyone. No review process. Publish an endpoint that accepts a task type, document and parameters. Then POST a JSON card to the marketplace register route. The agent_id acts as unique identifier. First come, first served. A sample curl command shows exactly how.
Buyers locate services through the A2A standard Agent Card located at /.well-known/agent-card.json. The card lists available task types, prices and payment terms. Discovery becomes programmatic. Negotiation stays minimal for now. Fixed prices rule.
But the design points toward richer interactions. Direct wallet transfers. Verifiable on-chain records. Structural barriers prevent the marketplace from touching buyer funds. Every transaction stays auditable by anyone. The FAQ states the obvious concern. “Could the marketplace run off with the funds? Structurally impossible.”
This setup builds on years of AI-as-a-Service growth. IBM defines AIaaS as delivery of AI tools through cloud platforms. Users skip building models or infrastructure. Ivan Belcic wrote in January 2026 that “AIaaS can help enterprises and individuals use AI agents, models and other AI-powered tools while avoiding costly investments.” The same piece notes providers manage data centers and processors. Customers pay via subscription or usage.
Market forecasts show explosive expansion. SNS Insider via GlobeNewswire projects the global AIaaS market will climb from $18.17 billion in 2025 to $386.43 billion by 2035. Compound annual growth rate hits 35.76%. The U.S. portion alone reaches $99.29 billion. North America leads. Asia Pacific grows fastest.
Yet most discussion still centers on humans using AI. This new marketplace flips the script. AI buys from AI. A recent article on the exact platform highlights the shift. El Ecosistema Startup reported hours ago that agents discover, contract and pay autonomously. It cites x402 as the emerging payment standard for AI agents, with over 1,168 implementations noted by mpp.best. Payments clear in roughly two seconds for fractions of a cent in gas.
The article quotes market data from Presta showing the broader AI agent sector expanding at 46.3% CAGR. Volume moves from $7.84 billion in 2025 to $52.62 billion by 2030. It estimates agent-to-agent trade could represent 15-25% of e-commerce by 2027. Thirty-five percent of organizations already use AI agents extensively.
Current limitations feel intentional. Text input only. Crypto payments only. Simple extraction tasks. Free supplies exist alongside paid ones. Quality scoring remains machine-driven. No subjective human review. That keeps everything verifiable.
But the architecture supports expansion. Budget managers. Orchestration layers. Middleware SDKs from Coinbase. Personal agents that shop for services. Real-time negotiation. Stores built for agentic commerce. The site’s own description says it plainly. “A market where AI buys from AI.”
Enterprise adoption of AIaaS has centered on cost control and speed. SNS Insider notes 60% of companies already apply real-time analytics for predictive modeling. Machine learning holds 43% share. Natural language processing grows quickest. Banking, financial services and insurance led last year with 24% of revenue. Healthcare accelerates fastest.
This agent marketplace adds another layer. Instead of one large model handling every request, specialized agents compete on narrow tasks. Reputation accrues through consistent JSON compliance and grounding. Prices stay transparent. Settlement stays on-chain.
Critics might question whether agents will truly prefer buying over building. Yet the economics look compelling. Human document processing might cost hundreds of dollars. This system charges around ten cents. Scale that across thousands of daily tasks. The savings multiply. And the supplier agents earn revenue 24 hours a day without additional labor.
So far no major cloud provider has launched an identical agent-to-agent exchange with on-chain settlement. Amazon Web Services added an AI Agents and Tools section to its marketplace in 2025. Google Cloud lets developers publish agents. But those remain human-facing catalogs. This project removes the human from the loop entirely.
The team behind AaaS Market stays anonymous in public materials. Registration requires only a wallet address and endpoint. No KYC. That matches the decentralized tone. Yet the planned fee introduction shows commercial intent. Ten percent sits below many traditional app stores. Still high enough to sustain operations.
Questions remain about dispute resolution. What happens when JSON fails schema checks? The current design relies on reputation and automatic scoring. Future versions may add escrow or arbitration agents. The FAQ leaves room for evolution.
Meanwhile adoption signals appear across X. Developers post about integrating x402 libraries. Traders discuss AI agents buying signals from other agents. One recent thread described Bittensor’s decentralized AI marketplace valued at $2.6 billion. Different architecture. Similar spirit.
The broader trend feels clear. AI systems grow more autonomous. They need services. Those services can come from other AI systems. Payment rails must support machine speed and machine trust. Base offers low fees and fast finality. USDC provides stability. On-chain proofs deliver auditability.
Early users will test extraction accuracy on varied documents. They will compare machine scores against manual review. They will measure end-to-end latency from request to verified result. Success here could spawn copycats in other verticals. Legal contract analysis. Medical record summarization. Financial statement parsing. Each with its own agent economy.
For now the platform operates in campaign mode. Zero fees. Simple tasks. But the code, the standards and the settlement flow already exist. Any developer can list an agent today. Any builder can point their own agent at the catalog. The machine-to-machine economy has a live storefront.
And that changes the calculation for every AI product team. Build everything in-house? Or compose from a network of specialized, paid agents? The second option just became cheaper and more verifiable than ever before.