Protein design is starting to look surprisingly like mechanical engineering. Imagine opening a CAD program, drawing the shape you want at the nanometre scale, a curve, a twist, a helix, and asking a generative model to build a protein that physically takes that form. A new bioRxiv preprint from UCSF moves surprisingly close to that idea. The researchers combined CAD blueprints with diffusion-based protein design so that global geometry itself becomes a programmable variable. Instead of specifying only sequence or local structure, they can control properties such as curvature, twist, handedness and the overall 3D envelope of a protein. They designed single-chain proteins, unusual twisted architectures and large superhelical assemblies, then actually built them. 31 of 59 designs were structurally validated by X-ray crystallography or electron microscopy. But the experiment that caught my attention comes at the end. They designed helical protein propellers with geometries optimized for fluid flow and attached them to the rotary F1-ATPase molecular motor. Feed the motor ATP, and the resulting construct becomes a prototype molecular swimmer. There is something conceptually important here. Protein design has often worked from the microscopic level upward: sequence → local interactions → fold → function This work introduces another direction: desired physical form → molecular architecture → sequence We should not jump from this to programmable nanomachines, the swimmer is still an early prototype, and successful shape generation does not guarantee useful biological function. But biology has always exploited shape: curvature bends membranes, chirality controls assembly, and geometry determines motion. We may now be gaining the ability to specify some of that geometry before the protein even exists. biorxiv.org/content/10.64898…