On the same day that OpenAI's new model overcame mathematical challenges, an MIT professor claimed that AI has the ability to transform matter like software programming Just now, MIT engineering professor Markus J. Buehler posted a tweet that explicitly echoes the latest developments in OpenAI: "If physical systems can be formalized as composable mathematics, we can point the AI that has been proven to solve long-term open mathematical problems towards matter itself. ” A few months ago, the Buehler team published research on programmable materials in the Journal of Solid Mechanics and Physics, and open-source the complete code and G-code. But the tweet further explains the natural extension of "AI conquering mathematics" to the concept of "AI compiling matter" by borrowing the latest achievements of OpenAI. Taking Pinecone as an example: Can we compile matter like compiling code, from observing biological hierarchical structures, to designing new active materials, to manufacturing and testing, to achieve a true end-to-end process? The team adopts a category theory approach: modeling each scale (fiber → layer → tissue → unit → organ) as a "module" with clear states, stimuli (humidity or heat), dynamic laws, and interfaces. The mapping between scales must strictly maintain the consistency of stimulus response dynamics: evolve first and then map, or map first and then evolve, and the results must be the same. This condition still holds under combination, and locally correct interfaces are still reliable when assembled into a complete hierarchy. Subsequently, they mapped biological structures to engineering systems using "implementation functors", translated them into validated manufacturing specifications, and directly compiled them into G-code executable by 3D printers. Finally, four types of actuators were printed: humidity bending, thermal bending, humidity torsion, and thermal torsion. The most amazing thing is' thermal torsion '- it does not require re derivation, it simply combines a validated thermal stimulation module with a torsion module. The generated G-code directly generates the expected motion, and the experimental error is within one standard deviation. This is the first time that a formal combination of multi-scale models has been brought from the biological level all the way to physical testing products. Category theory no longer remains in abstract mathematics, but has become a real object on a printing bed. For AI science, this is equivalent to providing a "type system" of physical perception for generative models - proposals can be rejected by interface checks before simulation and manufacturing; For engineering, the design space begins to expand with the validation component library, rather than linearly increasing with the number of cases. According to the core logic of the tweet, atomic level programming is entirely possible in the future. Because logic is scale independent: as long as the system can be formalized as composable mathematical objects and morphisms, and as long as interfaces can be rigorously checked, AI can reason and compile matter at smaller scales like solving open mathematical problems. The real obstacles are computational costs, quantum randomness, and atomic precision manufacturing, rather than the principles themselves. The tweet showcases a new paradigm: when AI can solve mathematical problems that haven't been solved for a decade, pointing it at matter itself may just be a matter of time. Pinecone is just the starting point, the programmable physical world may have just opened its first door.