AI Generated Tests as Ceremony (blog.ploeh.dk)

In a thought-provoking exploration, Mark Seemann critiques the growing trend of utilizing large language models (LLMs) to generate automated tests in software development. He argues that while LLMs can streamline the coding process, relying on them for test creation may lead to superficial results, where tests become mere "ceremony" without proper engagement or understanding. Seemann emphasizes the importance of epistemological soundness in software testing, highlighting the necessity of seeing tests fail to affirm their validity—a crucial aspect often overlooked when using LLM-generated code. Seemann raises fundamental questions about the reliability of LLM-generated tests, cautioning against the misconception that all passing tests guarantee correctness. He suggests that developers may fall into a false sense of security, as automated testing could devolve into a performative exercise lacking real rigor. Instead, he proposes that adopting practices like empirical Characterization Testing or even reversing the LLM's role—writing tests first and allowing LLMs to implement the system—could foster a more robust testing culture. Ultimately, while LLMs are set to play a significant role in the future of software development, the community must remain vigilant in maintaining the integrity of testing processes.