Original Research

Engineering papers on agent architecture, AI cognition, and systems that don't exist in any textbook yet.

Paper 003//14 min read

The Reverse Mirror: AI Is the First System That Helps You Reverse Engineer Itself

Reverse engineering has five levels. Most engineers stop at level one. The principles that matter, the ones that will define how the next decade of agent engineering gets built, all live at level five. This is the method that produces them.

reverse engineeringemergent behaviorAI cognitionprinciple extraction
Paper 002//11 min read

The Testing Disease: Why Most AI-Generated Tests Are Theater

AI can write a test suite with 94% coverage that catches zero bugs. The problem isn't the AI. It's the direction of flow. Tests written from source code are tautologies. Tests written from specifications are antibodies. And mutation testing is the only honest scoreboard.

AI-assisted testingmutation testingspecification-first testingtest tautology
Paper 001//12 min read

The Landscape: Emergent Personality via Environmental Navigation in Conversational AI

Every AI company is writing better system prompts. They're solving the wrong problem. Personality isn't content. It's navigation.

agent architectureconversational AIvoice agentsAI personality