Researchers introduced HARMONY, a framework for 3D scene reconstruction from a single image. It combines agentic reasoning for spatial understanding with visual geometry for alignment. The system first calibrates the camera and establishes a spatial frame. It then recovers room layout and object placement order using a vision-language model. Objects are placed hierarchically, from walls to furniture to decorations. Depth-first traversal conditions each placement on prior structure. A feedback loop refines geometry to match the input image. HARMONY outperforms baselines on synthetic and real-world indoor scenes. Qualitative tests show better object arrangements than GPT-6 Astra. The paper and project page are available on arXiv.