Same Tool, Different Brains — Why LLM Inconsistency Drives Me Nuts

Same prompt, same tool (opencode), same task — but different models give completely different behaviors. Why this happens and why it breaks reproducibility in AI-assisted development.