100%, and a lot of them are truly terrible use cases for LLMs.
For example, using a LLM to transform structured data into JSON, and doing it with two LLMs in parallel to try to catch the inevitable failures, instead of just writing code that outputs JSON.
