The field of scientific discovery is rapidly advancing with the integration of large language models (LLMs) and agentic optimization. Researchers are leveraging LLMs to automate the development of computational methods, enabling the efficient analysis of complex experimental data. This automation is leading to breakthroughs in various domains, including geospatial modeling, genetics, and physics. The use of agentic optimization is allowing for the discovery of new algorithms and models, which are outperforming state-of-the-art expert methods. Notable papers in this area include GeoEvolve, which introduces a multi-agent LLM framework for automated geospatial model discovery, and TusoAI, which develops and optimizes computational methods for scientific tasks. SciExplorer is also noteworthy, as it enables free-form exploration of physical systems using LLMs. Overall, these advancements are opening new opportunities for trustworthy and efficient AI-for-Science discovery.