Imagine effortlessly orchestrating complex workflows across a vast landscape of web-scale APIs using only natural language. This dream is becoming a reality thanks to advancements in Natural Language to Domain Specific Language (NL2DSL) generation. Researchers are tackling the challenges of converting everyday language into the precise code needed to automate tasks involving thousands of APIs. Traditional code generation methods often falter when faced with the unique names and sheer number of APIs involved, leading to errors and 'hallucinations' where the AI generates nonsensical code. This research explores optimizing Retrieval Augmented Generation (RAG) techniques, which involve grounding the AI model with relevant code examples. The core idea is to provide the AI with a context of existing code snippets similar to the desired task, along with detailed descriptions of the APIs involved, empowering it to generate more accurate and reliable DSL code. The study compares various approaches to grounding, using a fine-tuned model as a benchmark. The results reveal that while fine-tuned models excel in certain areas, the optimized RAG approach performs comparably for familiar APIs and significantly outperforms the fine-tuned model when dealing with unfamiliar or 'out-of-domain' APIs. This adaptability makes RAG incredibly valuable in real-world scenarios where the API landscape is constantly evolving. The implications are vast, ranging from automating business processes like invoice processing and sales lead integration to more complex orchestration tasks. The ability to plan and execute these workflows using natural language promises to democratize access to powerful automation tools, ushering in a new era of efficiency and innovation.