Third Place / AI Research

AI Systems Research · ActionPace

Emergency Room Systems Architect — AI App-Gone-Bad CPR

Structural signals in
composable AI systems

Research into the architectural gaps between what AI execution frameworks solve and what they leave undeclared — the implicit contracts that accumulate and transfer complexity onto the human operator.

cs.SE · cs.AI · arXiv forthcoming Working Paper
Toward a Control Architecture for
Composable AI Execution Systems
An empirical survey of six production AI execution frameworks using ensemble divergence methodology identifies eight architectural strengths and four anti-patterns. The survey finds no framework that has integrated control discipline at the architectural level. A theoretical model specifying a control discipline layer is developed from the findings.
Frameworks surveyed
6
AutoGen, CrewAI, LangGraph, n8n, OpenClaw, PydanticAI
Files analyzed
1,269
Three-phase funnel · $6.58 total analysis cost
Patterns identified
8 + 4
Eight strengths · four anti-patterns · confirmed at 2–9×