2025 Week 15

If you follow my note from last week, I mentioned that I started coding a new idea which I'm doing mostly by vibe coding. I still did a lot of that this week and I think I have some opinions to share about coding this way.

Generally I think it's been a bitter sweet experience. I'm happy I have something to help me bootstrap this with little nudging. But at the same time, I feel there is a lot of time wasted on trying to move the LLM in a certain direction. Precision has also being a big challenge, the tendency, that your feature might not work or that how something looks will change, the more edits are made is non-zero and I had to learn look out for this and even instruct to focus on very specific files.

Since I know the technologies being used very well, I also find my self questioning a lot of the implementation where I think it can potentially lead to issues.

Now, the big one was when I got to a point where no matter the amount of prompt I did, the problem wasn't going to be solved. At this point, I had to resort to doing it myself. Thankfully, I had followed up with every aspect of the code generated and was able to know what I need to do.

Given that I'm also using this to build an AI app that will be doing relatively the same thing, I can't but wonder if I will be able to solve some of this problems. I don't really worry about this since for me, it's all about having the basic thing up and running and then build on top of that. I believe there's some lessons that can be gained from using these ones while I make mine too.

So, these are my experiences so far. I'll still be vibe coding this until I reach the end of it and release to the public.

---

Till next time, Bosun