For a long time, I thought using AI was mostly about taking the things I already do and doing them faster and with great accuracy. And that's still the case for a lot of what I do.
But recent work showed me the real shift happens when you hit a wall in your own skill set.
Every month, I consolidate data from disparate sources. Each file arrives with its own quirks and formatting. My go-to tool has always been Power Query, and I'm pretty good at it.
But every time a source file changes, a column moves, a folder is renamed, I add another file with a different layout or I move the location, it's another half-hour to make the PQ change and then another half-hour to double and triple check it. After all, the output has to be correct.
The truth is, building a routine resilient enough to handle those constant shifts requires a level of coding experience. So it wasn't an issue of not enough time, it was a gap in my capabilities. I didn't want to admit it, but I knew it. Spending another hour or so hunting to find the next solution to fix my specific issue, wasn't going to move me on much.
So I tried giving the raw, disparate files directly to Claude. Instead of forcing my own limited code to adapt, the LLM processed the structural changes, normalized the inputs, and produced a great-looking output that I knew was accurate.
The real value wasn't in speeding up my existing workflow. It was in handing off a task that required skills I simply don't have, and letting the LLM bridge that gap. And you know what, because the LLM explains what it's doing, my skills-set and understanding of what goes on 'under the hood' is increasing.