Do you let LLMs write both your tests and your code? (lobste.rs)

A developer is contemplating a guiding principle for their team regarding the use of Large Language Models (LLMs) in software development, specifically whether to allow them to write both tests and code. The intention is to establish a rule that limits LLM contributions to one domain at a time, as recent experiences have shown that when LLMs are permitted to generate both simultaneously, the results can lead to overly complex and redundant code that prioritizes test generation over functional clarity. This discussion highlights a significant concern in the AI/ML community regarding the practical applications of LLMs in software engineering. The emphasis on maintaining a clear separation between code and test generation could help mitigate issues of overengineering and tautology, ultimately improving software quality. As teams increasingly integrate LLMs into their workflows, understanding the implications of their dual capabilities may be crucial for ensuring efficient and effective software development processes.