Do you let LLMs write both your tests and your code?
A Lobsters discussion thread opened by travisgriggs asks the community whether they let LLMs write both their tests and their code, and what stance or experience others have had. The author says they are toying with adding a rule to their values.md that the LLM can only significantly generate one or the other. The prompt for the question is a merge request they are currently reviewing, where it is clear the LLM mostly did both and the tests are highly tautological. The author also says they can see how the LLM over engineered the code just to be able to write more tests. The thread drew one comment at the time captured. Commenter tomleo says they sometimes use LLMs for code and unit tests, then write the integration and e2e tests themselves, adding that it depends on the code and domain. The exchange stays at the level of team practice and review experience: no specific tool, model, or policy outcome is named beyond the proposed values.md rule. The concrete observations offered are the tautological tests and the over engineered code in the merge request under review, and the split approach described by the commenter.
The thread surfaces a review concern that LLM-generated tests and code together can be tautological, which developers may weigh when setting team rules.