Software engineers are grappling with the increasing quality of AI tools for code generation, particularly large language models (LLMs). Many view code as becoming "cheap," akin to fast fashion clothing—quickly produced and often of low quality. For engineers focused on long-term software systems, this raises questions about maintaining quality and managing risks. The challenge extends to hiring and training new developers, as the landscape of coding evolves. In contrast, data scientists see value in these tools, especially when their work often involves creating ad hoc solutions. Many professionals discard code after a single use, making the notion of disposable code less concerning. The article highlights two main uses for LLMs: generating useful code for data tasks and the pitfalls of relying on LLMs to create synthetic data. The latter is problematic, especially in user experience research, where real human data is irreplaceable. LLMs can streamline the creation of data extraction tools, such as converting poorly formatted PDFs into usable data. This capability allows data professionals to access previously untapped information with less effort. However, the downside is that LLMs can produce unreliable code, complicating the evaluation of the generated data. Data extraction often involves unpredictable factors that aren't easily tested, making it challenging to trust the results. The article notes a shift in the need for software engineering assistance. As tools for code generation and understanding improve, data professionals find themselves more capable of handling low-risk coding tasks independently. This change could lead to a more efficient workflow, as barriers between data work and engineering resources diminish.