Large language models (LLMs) have shown promise in various fields, but can they handle the complexities of scientific workflows? New research explores how LLMs perform in configuring, explaining, translating, and even generating scientific workflows. The results reveal a mixed bag: while LLMs excel at explaining existing workflows, they struggle with tasks requiring deeper scientific knowledge, like configuring new workflows or translating code between different workflow systems. This is primarily due to LLMs' lack of specific training on the nuances of scientific computing. However, the research also offers a glimmer of hope. By providing LLMs with more context, such as example configuration files or code snippets, their performance improves significantly. This suggests that with targeted training and clever prompting, LLMs could become valuable assistants for scientists, automating tedious tasks and making complex workflows more accessible. The research also highlights the need for specialized benchmarks to accurately evaluate LLM capabilities in scientific domains and guide the development of more powerful, scientifically-aware LLMs in the future. Imagine a future where scientists can simply describe their desired workflow in plain English, and an LLM automatically generates the necessary configuration files, code, and even benchmarks – this research takes a crucial step towards realizing that vision.