This paper introduces a framework that utilizes large language model (LLM) agents to extract and convert mathematical models from engineering literature into executable code. Autonomous or semi-autonomous conversion of literature into code facilitates downstream tasks such as hypothesis generation, verification, and benchmarking. Focusing on heat exchanger design, our approach efficiently integrates model extraction, code generation, and performance optimization, with minimal human intervention. The system's knowledge base is continuously refined with each new paper, leading to ongoing improvements. Experiments conducted on 115 research articles using the HxAgent approach demonstrate substantial improvements over the previous non-agentic baseline, HxLLM. Although the work is still in progress, the results highlight the potential of agent-driven workflows in advancing scientific discovery.