GLM 5.2 vs Composer 2.5, the cheap fight

Ray Fernando demonstrates a live coding session using AI-driven orchestration to build a real-time voice transcription and translation app for macOS and iOS. He advocates for "Wave engineering," a multi-step orchestration technique that replaces inefficient looping by fanning out tasks to parallel agents, aggregating results, and verifying outputs through a multi-model approach. By utilizing Factory AI’s "Missions" platform, he delegates coding tasks to the GLM 5.2 model, uses GPT 5.5 for validation, and relies on Opus 4.8 for high-level orchestration. This workflow optimizes token usage and improves system reliability by treating AI-generated code as a claim that requires rigorous validation. Beyond technical implementation, the session emphasizes the importance of community-driven knowledge sharing and the practical application of agentic workflows to accelerate software development cycles.