Designers read images quickly. A top view, screenshot, or rendering can reveal problems faster than text: the proportions are wrong, the entrance is too narrow, the furniture is crowded, the wall direction is strange, or the lighting emphasis is off. That makes visual feedback an important part of an AI design workbench. But visual feedback can also become a new source of disorder. If it stops at "this looks wrong," or if AI only patches the current picture, the project quickly fills with screenshots, opinions, and one-off fixes. The key question is: how does visual feedback return to the structured design project? The image is not the final truth Screenshots and renderings are useful, but they are not the only source of truth for a design project. An image can show what appears to be happening, but it usually cannot answer: does this problem come from spatial dimensions or camera perspective? is the model wrong, or is the rendering expression wrong? should the material change, or should the lighting change?…