Over Tooling

I like to think that I have a systems oriented brain. I don't like solving instances of a problem, instead I prefer to identify the root cause, and solve that instead. This is a good instinct for a software engineer to have. However, it can be terrible in other areas of life, even at work.

This has only gotten worse with the rise of agentic programming. In years past, if I had a problem the first step was always a google search. If I found a solution within a minute or two, I just did the thing and moved on. I figured the answer would always be out there, so why capture it. If the answer took a little longer to grab, I'd add it to a notes file or wrap the command in a bash script/alias. Very rarely, I would write my own script to solve the problem.

Today, I and many other engineers I know treat claude code as a replacement for google. But, and this is an important distinction, we aren't looking for solutions that are out there. We are making our own. We are using claude to write solutions for things that already exist, and can be maintained by people better than we ever could. Two recent examples:

  1. An engineer on my team prefers viewing code for review on his local machine instead of the github interface. Claude wrote him a multistep script that fetches the PR, grabs the branch, and checks it out locally. Had he googled, he would have found gh pull checkout XXX which does the same.
  2. I am also having some frustration with the github UI. I went many, many steps further than my engineer and started working on my own TUI for interacting with all parts of github that turned out to just be a huge collection of scripts that probable should have been named aliases.

Both of these things were a great way to practice working with Claude. But they weren't the easiest way to solve a problem, they were expensive to produce, they are now things that have to be maintained, and most importantly I am never going to know what is out there if I just have an LLM "solve" my every need. So I am going to try to focus back on short scripts and using solutions that are out there before reinventing the wheel.