Argon is Greek for "lazy." Fitting name for a model that just skipped 15 years of Google's hardest work. 2007: Street View launches in five cities, built on ~125 million photos shot from VW Beetles. 2007: Hamburg becomes Google Earth's first 3D city, stitched together from about 1,000 aerial photos taken 500 meters up. 2013: five million miles driven across 47 countries with laser rigs bolted to the roof. $75M+ spent, by one reseller's estimate. 2018 and 2021: Project Tango and Poly, the "cheaper" paths, both get shut down. October 13: Gemini 4 Argon looks at a few photos of a desk and turns it into a playable racetrack. The demo is pure aura farming, but Nano Banana's model card still lists 3D reasoning as a limitation. The image model isn't doing the geometry at all. Turning -v on for a sec: Argon writes Python for headless Blender over MCP, collision-checks it, lets Nano Banana texture it, then hands physics and engine audio off to Three.js. I see the same thing in the drug discovery agents I keep tabs on, ChemCrow, DrugAgent, the autoresearch loops. The LLM doesn't do the chemistry, it orchestrates the tools that do. AI as the default actor, humans only showing up for the last 5%. Space stopped being something you scan and turned into something you write code for. Street View needed a whole fleet of cars, Argon runs on $2 per million input tokens. Everyone's a gangster until the desk has to become a real building with real tolerances... baby steps bro