Architecture Guardrails for AI Code Agents (ANMA)

A YAML-based contract system that enforces architectural rules on cheaper LLM coding agents, making them follow guidelines consistently. Target: engineering teams using cost-effective AI models for code generation who need reliability.

Agentic coding workflows are proliferating fast — Cursor, Copilot, and a wave of cheaper models like DeepSeek and Qwen are pushing teams toward multi-agent pipelines where architectural drift becomes a real ops problem, not a theoretical one. No clear incumbent owns the "guardrails as code" layer specifically for agentic code generation, though Cursor Rules and custom system prompts are the current DIY substitute most teams cobble together. The $1k–5k/mo band is plausible for small engineering teams paying per-seat, but it assumes they'll pay separately for this rather than expecting it bundled into whatever agent orchestration tool they're already using. The biggest risk is that the major platforms — Cursor, GitHub Copilot, or agent frameworks like LangGraph — absorb this natively, making a standalone YAML contract system redundant before it gets traction.