Fedus explains that while ChatGPT was initially too weak for materials science, recent advances in models, reasoning, and agents have made it possible to apply AI to physical experiments. Periodic Labs uses language models as an orchestration layer to direct experiments and coordinate specialized neural networks for atomic systems, creating an integrated closed-loop system that combines literature, experimental data, and multiple modalities. The company aims to provide an intelligence layer for companies facing materials and process engineering bottlenecks, with a business model comparable to biotech. Fedus envisions accelerating physical world development through AI-driven atomic rearrangement synthesis, potentially matching digital progress in productivity for semiconductors, energy, and aerospace. He highlights the multidisciplinary collaboration between physicists, chemists, AI researchers, and engineers as key to overcoming data bottlenecks through larger-scale experiments and automation. Regarding AGI, Fedus notes that intelligence is not scalar and that software engineering self-improvement is happening now due to verifiability, while AI research progresses more slowly due to longer experiments.