Jive Coding with ChatGPT

That’s right, jive coding. Not vibe coding. Why? Two reasons. First, “jive” is more fitting for those of us who were born in the 1970s. I wasn’t old enough to leave the house on my own in the 1970s, but I was born in the disco decade.

More importantly, “jive” is likely an apt descriptor of the coding produced when someone with little background in a technology creates something with a heavy dependency on generative AI. As the Bee Gees have taught us, jive talking is largely fallacious or non-sensical, and I’d say this aptly describes most scenarios where an AI user believes he or she is learning to code by having ChatGPT do the work.

But caveat: I did say, “most.”

This weekend I started a jive coding session for a pet project with my favorite AI assistant. I want to design a website that collects data from users and allows them to search the resulting dataset.

Keep in mind, my background is, well, backend. I’ve worked for decades in data, not in UI or web development. I have some programming background, but it’s been years since I touched C#, and my more recent coding has been Arduino and 3D modeling with OpenSCAD — very subject specific.

My project is going to need a lot of AI vibe jive!

The first tasks included setting up an environment where I could work with data and create a website, load and clean a public dataset of business locations, then get to the website coding.

ChatGPT Pro already has a thorough picture of my skillset from our past interactions, so I started by creating a new project and supplying a very detailed description of my project goal. That’s an important key: AI is much better at assisting when it knows your vision from start to finish.

I described the must-haves for the project as well as the nice-to-haves and opportunities for scaling later. ChatGPT helped me develop a plan, including finding a host system sufficient for all the infrastructure I need. About thirty minutes into the project, I had a $6/month platform set up on a reliable hosting service, with MySQL, Python, and other tools installed.

And then the serious jiving started.

The weekend tasks included two major instances of jive coding. First, I moved my dataset into MySQL and started cleaning data. Second, my data included map coordinates, but not addresses. I found a nice website service to convert coords to addresses via an API, but I could only send one set of coordinates per second through the API, and I had 80,000 records to process. Sounds like a job for Python.

Data cleanup first. I’ve spent decades in Microsoft SQL, but hadn’t touched MySQL until this weekend, so I leaned on ChatGPT to give me the MySQL syntax for dozens of queries.

Conclusion: because of my expertise in MS SQL and the significant similarities between it and MySQL, my jive coding with ChatGPT was actually an effective learning experience. I’m already writing MySQL queries without syntax references, so I’ve clearly retained knowledge from the session.

However, Python is a different story. My comparable programming experience isn’t as similar as MS SQL and MySQL, and far more rusty.

ChatGPT came up with a great Python script for sending my data records to the API, one per second, and writing the results back to MySQL. I can easily comprehend the script, too. I see how the database connection string is created, the data cursor is defined, and how the API is constructed.

But could I sit down and write this script in Python tomorrow without using ChatGPT or referencing significantly other help? Absolutely not. Jive coding gave me a functional script that worked for my use case, but it was an example, not a lesson. ChatGPT gave me a fish, it didn’t teach me how to fish for myself.

This isn’t necessarily a problem. If I just want to accomplish an immediate task, mission accomplished! If I want to learn how to write Python effectively myself, I know how to find resources and learn.

As many other bloggers have pointed out, though, there are potential problems when relying on this jive approach. Whenever you build something public-facing, you have to ensure strong security — and if you’re just jiving, you have no idea whether your AI assistant implemented security correctly.

Keep that in mind when implementing AI in your own projects or enterprise environment. Too much dependency on jive coding could be as embarrassing as practicing your Saturday Night Fever moves with your webcam on.