I think a big part is the misperception that it’s “easier” and less effort to run stuff through LLM than to design an effective deterministic process.

Would love to know how you’ve managed to counter this as the drive to throw everything at LLMs is driving me insane.