Jive Coding, Part 3

My jive coding web app is complete! At least, the v1 is complete, because as Excel has taught us, old software never dies, it just keeps iterating. Check out my work at ISleptOkay.com.

Let’s talk best practices. As I mentioned in Part 2, my slightly sarcastic relabeling of “vibe coding” to “jive coding” is exactly that – slightly sarcastic.

On the one hand, I’m referring to this practice as “jive” coding because you’re not truly training, you’re just following instructions without understanding what you’re doing.

On the other, I’m pleasantly surprised at how much I HAVE learned through the experience.

Think of vibe/jive coding as putting together a room full of Ikea furniture. A great deal of the time, you’re hoping that your interpretation of the pictures is correct. Invariably, some of those pieces aren’t going to go together, the shelves are going to lean, the frame isn’t going to line up, or the doors aren’t going to fit.

Those moments are when the greatest learning experiences happen. In the case of jive coding, though, you have a some great resources to help you puzzle through – ChatGPT, Claude, etc. (In the case of Ikea constructions, you get your neighbor, Jim, who hasn’t met a problem that can’t be fixed with a nail gun.)

Here are my top tips for both a successful project and some possibility of learning:

• Start with business requirements. The AI helper is going to create technical content for you, and help you find technical requirements. (In many instances, ChatGPT would ask questions like, “Do you want me to show you how to harden the database access rules,” etc.)
  
• Use git to maintain branches. This is especially useful for feeding back information to the AI. For example, when ChatGPT gives you a code snippet to drop into a file, it’s wise to feed it the current entire contents of that file so the AI can ensure the new snippet aligns correctly.
  
• Build the master blueprint in another document and constantly feed it back to your AI. My hotel website included a couple dozen files, and ChatGPT had a tendency to say, “If you have this kind of file…” A blueprint is very helpful for reminding the AI of the architecture it’s built.
  
• Learn. Review the code that you’re cutting and pasting, and ask the AI to explain it. Not only will you build some familiarity in general, when things go wrong you’ll have a better idea of where the problem is, and you may remember things the AI didn’t put into memory.
  
• Note the questions at the ends of the prompts. I like that ChatGPT ends each step with a question like, “Would you like to do <this> next?” Quite often, though, the current step didn’t just fall into place correctly, and it could take a dozen or more queries to get things straightened out. When that happens, you’ll want to remember that the AI made a suggestion for the next logical step.

More on this project and the whole jive coding experience to come!