One thing I can't figure out how to get LLMs to do is truely finish work. For example if I have 100 items that need xyz done to them, it will do it for the first 10 or so and say ~"and so on". I have a lot of trouble getting LLMs to do tasks that might take 10 mins - 1h. They always seem to simply want to give an example. Batch processing is the answer, I guess, or perhaps more 'agentic' models/tools - but I wonder if there are other ways.