I Used Generative AI To Write Python and It Was Delightful*

Elliot

• 3 min read
I Used Generative AI To Write Python and It Was Delightful*
Photo by Maxime VALCARCE / Unsplash

I admit I wonder how creative I actually can be. As much as I am fascinated by the potential of AI, I struggle to think about practical applications that lead to interesting ideas and applications. Forget about founder-sized ideas, I mean day-to-day stuff. This week I finally had one of those aha moments. It may sound small to many but since I am actively developing this skill, I am taking my win and talking about it.

I've been working on developing some of my purple team skills and found a book that included a link to their website which hosted the book's lab files.

The files URL was basic auth protected, which I knew, because the book includes the username and password.

When I got there, I was surprised to discover there were over 30 files, but no single compressed file collection:

There was no way I was going to right click ALL of these files and download them all. I don't have a download manager, and I didn't want to use a service or download anything to get these files. So I decided to "write" a one off Python script to do this. My Python is terrible and today I was not willing to take the extra time to learn Python syntax, so I qualify "write" because - and here's the aha moment - I was going to have Claude write it for me. I was going to tell Claude what to do, but I was going to try to write as little Python as possible, with the goal being writing no Python at all. Could Claude do it for me in its entirety?

That isn't worth writing about; of course Claude did it easily. Claude and other LLMs have been doing this for years already. The script took 20 minutes to complete and it worked exactly the way I asked.

I hung around on this idea - that it worked exactly the way I asked - and wondered if my successful use of it still was connected to me as an individual:

  1. My previous experience. I was able to articulate my requirements clearly because I understood the process I wanted Claude to work with. I could describe the auth model and other relevant characteristics, and the output reflected what I wanted because I knew what to ask for. Claude can do this much faster than I can write code. And I can write what I want faster than I can write functioning code in a language I am not proficient in.
  2. My curiosity. I didn't want this to be a StackOverflow copy/pasta type situation - I was about to automate an interaction with a website and wanted to at least be polite about it. That meant at least digging into things that on the surface didn't look right. Before I ran it, I asked follow up questions, chased rabbit holes a little bit. I spent this time saved asking follow up questions and validating assumptions. I felt like my time developing was used more effectively.
  3. I admit to Claude my "reading" level. If I want something at a lower engineering lexile level, I will prompt early asking for a simple explanation. If it's a concept I understand well and just need a reminder, I will tailor the prompt appropriately.

The asterisk? It was delightful, and I am still 100% confident that the "human software engineer" will not be replaced anytime soon. It is a really effective pair programmer, though, one that will never get annoyed at my questions, will make mistakes like I will, and find them hopefully with 100% accuracy.

This is said broadly - I believe developers tend to prefer a builders mindset, and builders have tools like frameworks, languages, IDEs. This experience has reinforced to me that AI is a crucial tool in the developers toolkit and like any other tool I will get better at using it the more frequently I use it. I hadn't considered this before now.