AI agents build 3D scenes from photos but have no idea if they got it right

Researchers have introduced a method called LEGO-Anything that converts individual photographs into editable 3D scenes by generating Blender code. While GPT-6 Astra achieved a 53 percent reconstruction accuracy during testing, the study highlights a fundamental limitation in current AI agents. Across all models evaluated, the systems consistently failed to assess the geometric accuracy of their own outputs, suggesting that these agents cannot yet reliably verify the spatial consistency of the scenes they create.