The narrative describes the author's journey in leveraging large language models (LLMs), specifically GPT-4, to create a GitHub Action to Earthfile converter, aimed at easing the adoption of Earthly, a build automation tool. The author shares insights into overcoming challenges such as limited context windows and model hallucinations by employing various prompting techniques, including breaking down tasks into manageable steps, using chain-of-thought prompting, and providing corrective feedback to guide the model's learning. Despite the initial complexity, these methods allowed the author to achieve meaningful results and highlighted the potential of LLMs when combined with thoughtful instructional approaches. The article also emphasizes the benefits of Earthly in improving build reproducibility and efficiency in GitHub Actions, encouraging readers to explore its capabilities for enhanced build consistency and speed.