A glass hotel in Las Vegas melted the plastic lounge chairs by its pool and singed guests' hair. The Vdara's concave, south-facing facade acted as a giant parabolic mirror, focusing sunlight straight onto the pool deck. The same architect repeated it in London — the "Walkie-Talkie" at 20 Fenchurch Street, whose facade melted the bodywork of a Jaguar parked on the street below. Two buildings. The same physics failure. The same gap in the design process. What worries us as these tools spread through architecture and engineering: they make exotic curved geometry trivially easy to produce, but have no physics engine. They generate pixels, not load paths. These hazards are the dramatic end of a much bigger number: 80% of construction cost deviation comes from design changes, not construction — about $177B in rework a year. A ray-tracing pass during conceptual design would have caught the Vdara's "death ray" in milliseconds. The everyday version is quieter: a generative concept specs double-curved glass at $100-500 a square foot instead of flat tempered at $18-25, and nobody catches it until value engineering — ten ETABS re-runs and two weeks of senior-engineer time later. The fix isn't a better render. It's physics in the loop while the concept is still cheap to change. For the structural engineers here: what's the most expensive "beautiful" concept you've had to value-engineer back to reality — and how late did it get caught? #StructuralEngineering #GenerativeDesign #AEC