Handing an LLM a file and asking it to extract data out of it with no further context or explanation of what I'm looking for with good results does feel a bit like the future. I still do add context just to get more consistent results, but it's neat that LLMs handle fuzzy queries as well as they do.