@infowithbytes We did exactly this on the ops side - an agent that reads our queue and writes the daily status update. The LLM part was easy. The glue around it was the actual work, and nobody talks about that part.
@wadefoster The vendor lock angle is the obvious one, but the second is the one I see more - people defaulting to a model for steps that never needed one. Reran our agent pipeline on a cheaper open model last week, cost dropped a lot, and it was fine for what we needed.