I find there’s a tension at the moment when building tooling to support your day-to-day work as a software engineer. The tension arrises because coding agents are extremely good at “one-shotting” prototypes (and I think a lot of tooling need can be solved by protoypes) but after you have solved your problem, what do you do with the tool you are left with?
As an example, I recently needed to understand some information that was stored in a database. Instead of poking around the tables and trying to piece together the information in my head, I got Claude Code to “vibe” up a web app that displayed the records in a way that was way more digestable.
However, once I’ve used my tool to understand the information at hand, do I keep the tool? It was basically free and took basically no effort. If I have this problem again, I can just vibe up something similar. In fact, I could generate as many iterations on the intial idea as I want. How do I know this version of the tool is the best version of the tool? And then how do I know that next time I’m in this problem space that my needs for the tool aren’t different enough that keeping this version isn’t over-fitting?
But, I decide to keep my tool and now a question bubbles up, should I improve it? It solved my problem and I could move on but… maybe the layout of this bit of information could be made better, the way to invoke the tool doesn’t have great DX, and the contrast on this text makes it kinda hard to read. I’m now