Imagine a world where designing complex mechanical parts is as simple as describing what you need to an AI. Researchers are exploring this exciting possibility by leveraging the power of Large Language Models (LLMs) to create Mechanical Design Agents (MDAs). While LLMs excel at understanding and generating human language, they aren't inherently equipped to handle the intricacies of 3D modeling. This research introduces a novel approach to bridge this gap. Instead of directly asking an LLM to produce a CAD model, which often yields unpredictable results, they propose a step-by-step training process to guide the LLM's learning. This involves feeding the LLM foundational code representing typical part designs, extracting key features, and then analyzing these features in relation to real-world mechanical properties. Think of it like teaching an LLM the language of mechanical engineering. The LLM learns to associate natural language descriptions with specific design parameters and code sequences. Through iterative evaluation and refinement, the MDA gradually improves its ability to translate user requests into accurate 3D models. Experiments using this method show promising results, with the MDA successfully generating models for components like screws, gears, and even pneumatic joints based on simple text prompts. Moreover, the MDA allows for parameter customization, enabling users to fine-tune dimensions and constraints. While this research marks significant progress towards AI-driven mechanical design, it also highlights ongoing challenges. Building truly versatile MDAs capable of handling diverse design tasks requires further refinement and extensive learning. Still, the ability to generate complex 3D models from simple text descriptions offers a tantalizing glimpse into the future of engineering and design, where AI collaborators could significantly accelerate the development process.