Coordinating multiple large language model (LLM) agents requires solving the same fundamental problem as financial markets: how to allocate scarce computational resources among heterogeneous agents with incomplete information, without central authority, while maintaining quality and safety guarantees. We present MIAA (Market-Inspired Agentic Architecture), a coordination layer that maps financial market mechanisms to multi-agent orchestration. MIAA introduces six agent functions derived from market roles (Generator/Speculator, Router/Market-Maker, Validator/Hedger, Reconciler/Arbitrageur, Ledger/Clearing-House, Research/Analyst), a factual reputation system based on Exponential Moving Average and calibration tracking, a circuit breaker system with function-level, task-level, and system-level protection, and a constitutional layer (Inverse Constitutional AI) that blocks unethical actions proactively before execution. The architecture integrates with AGORA OS (semantic/ethical cognitive layer), SYMBIONT (bio-inspired adaptive layer), and Immune Middleware (defense layer) to form a Unified Organism. We describe the full architecture, the theoretical foundations of the market analogy, the implementation (approximately 1,600 lines of pure Python with zero external dependencies), and 20 integration scenarios ranging from obvious to disruptive. A test suite of 73 passing tests validates correctness. MIAA is available as an open Python library and planned as an MCP server for direct use in Claude Code sessions. Part of the Artisanal Intelligence Program — a Lakatosian research program on human-AI interaction.
Renato Aparecido Gomes (Mon,) studied this question.