Researchers at EPFL have developed a new artificial intelligence framework called Synthegy that uses large language models to assist chemists in designing complex molecules. Led by Philippe Schwaller and first-authored by Andres M. Bran, the study published in the journal Matter aims to solve two major challenges in modern chemistry: retrosynthesis and reaction mechanisms. Retrosynthesis involves working backward from a target molecule to identify viable synthetic pathways, while reaction mechanisms explain the electron movements behind chemical changes. Traditional computational tools often struggle with the strategic reasoning required for these tasks, frequently generating vast numbers of options without the chemical intuition needed to select the best ones. Synthegy addresses these limitations by positioning large language models as reasoning engines rather than direct generators of chemical structures. The framework combines established search algorithms with AI capable of interpreting natural language. This approach allows chemists to communicate their goals using plain English, such as requesting the early formation of a specific ring or avoiding unnecessary protecting groups. Once a request is made, traditional software generates numerous potential routes or mechanisms. Synthegy then translates these outputs into text, analyzes them against the user's strategic instructions, and scores each option. This process enables researchers to efficiently rank, filter, and understand candidate pathways based on specific criteria. In a double-blind expert study, 36 chemists evaluated 368 valid assessments of routes generated by Synthegy. The system's evaluations aligned with human judgments 71.2% of the time on average. The framework successfully detected unnecessary protective steps, assessed reaction feasibility, and prioritized efficient pathways. The study found that larger and more advanced language models demonstrated superior performance in analyzing chemistry at multiple levels, from individual functional groups to complete synthetic strategies. Andres M. Bran emphasized that the user interface is critical for adoption. Previous tools relied on cumbersome filters and rules, whereas Synthegy empowers chemists to iterate quickly through complex synthetic ideas simply by talking to the system. The technology effectively bridges the gap between synthesis planning and mechanistic insight, allowing scientists to express their goals in natural language and receive strategically relevant solutions. This innovation redefines how AI supports the chemical sciences. By enabling natural language interaction, Synthegy makes advanced computational tools more accessible and could significantly accelerate drug discovery and reaction design. The unified interface allows for the integration of expert hypotheses and reaction conditions as text, facilitating a more intuitive workflow. As the field evolves, this approach suggests that language models can serve as versatile evaluators, guiding traditional algorithms to solve complex chemical problems with a level of strategic insight previously difficult to automate. The researchers believe this methodology will foster new advancements in creating life-saving drugs and advanced materials by making the planning process faster and more aligned with expert intuition.