Researchers have developed a framework for agentic artificial intelligence that enables scientific discovery through the revision of representational regimes in materials science. The framework uses category theory to account for the process of discovery, which involves updating and refining existing knowledge while preserving provenance. Two systems, Builder/Breaker and CategoryScienceClaw, are presented as instantiations of this framework, demonstrating its application in protein-mechanics world modeling and typed skills-based knowledge-computation graph construction, respectively. The authors argue that category theory can serve both as a mathematical language for discovery and an engineering specification for self-revising AI systems. --- Why it matters: This work matters to researchers in AI because it provides a novel framework for agentic artificial intelligence that enables the revision of representational regimes in materials science, which could lead to more accurate and efficient scientific discoveries. The development of self-revising AI discovery systems has significant implications for fields such as materials science and engineering. Source: https://arxiv.org/abs/2606.01444