I Ran Three AI Agents from Three Different Companies on One Complex Project. Here's the Coordination Layer That Made It Work.

The Setup Nobody Writes About Most multi-agent AI articles describe a pipeline built on a single vendor's framework - GPT-4 calling GPT-4 in different roles, or a CrewAI setup where every agent is the same model wearing different hats. That's not what I did. Before I describe it: if you've seen this done elsewhere - three vendors, separate CLI sessions, git as the only coordination layer - I'd genuinely like to know. Drop it in the comments.