Day 1 of Vibecoding. Making a Journaling app

What is VibeCoding?

Imagine you’re building with LEGOs, right? Sometimes, you have a super clear picture in your head of exactly what you want to build – like a specific spaceship with all the right pieces. That’s like regular coding, where you plan everything out very carefully.

But sometimes, you just start playing with the LEGOs, putting pieces together that feel good, and seeing what cool thing comes out! You might not know what it’ll be when you start, but you just build because it’s fun and it feels right.

Vibecoding is like that! It’s when grown-ups who make computer stuff don’t plan every single little step super carefully. Instead, they just start writing the code that feels good, or exciting, or fun to them. They let their “vibe” or their feeling guide them, and see what awesome computer thing they create! It’s like playing with code to see what happens, instead of always having a strict plan.

VibeCoding a Journaling App

code me a journaling ai app in python. It takes in and stores daily and weekly entries. It has functions to anaalyze different times frames like weekly or monthly. can chat back as the writer feedback from chatgpt or gemeini

I put this prompt into 4 AI LLMs: Gemini, ChatGpt, Claude, and Vzero. Gemini and Chatpg are more general Large-Language Models (LLMS), while Claude and Vzero have more features and are more specifically geared toward Developers. Within 5 minutes of hitting enter, I was already faced with runnable code of the whole app.

Their solutions are all pretty similar. Notably the General LLMS took. amore hardcoded approach of analyzing the entries in chat using mood keywords, while Vzero actually put a Separate AI Layer ontop of the AI Journaling app in order to analyze entries with AI.

All four have decent explanations of the code. Though I do like how Gemini examines what you’ve done and suggests improvements, but i’m sure you can just prompt the LLMs for future suggestions and ask them to code them in.

I feel Vzero has the most robust solution, they use SQL instead of local json files, and implement AI with AI keys as a separate functional layer, so you can test out your base program at first, and then use the AI features on top of these.

I think any of these solutions would have taken me a few days of coding with my own skills, but in this case, I had 4 prototypes within 5 minutes. Structurally they are all the same. I can see the difference in approaches and slightly differing styles of code. I could probably use this as a learning activity and ask it to either generate the code in different languages from the prompts, or translate to different languages.