Struggles with Vibe Coding (Part 2)
"Waste" of two hours! One line change!
My personal Chinese app, the one I built to store vocabulary I don’t know, suddenly stopped saving data.
The weird part: I hadn’t touched the code in months. So when I just couldn’t add new entries, it was confusing and very annoying.
The fix was simple (basically a one-line text change), but getting there was a massive runaround. Took me two hours, lots of ranting, and yelling at bolt/chatGPT.
Debugging with Bolt
I started by asking Bolt to debug the issue. It suggested a bunch of possible causes: database issues, API formatting problems, frontend bugs and each time it would confidently say something like “Ah, I found the issue now.” And then… the bug would still be there 😡.
I went back and forth like this SO many times with Bolt.new, burning tokens and a lot of time. Eventually, ChatGPT coached me into utilizing the developer console and finally debugged it. “model not found.” Claude had updated its models, and the model my app was pointing to no longer existed. So every LLM call failed, which made the app look like it just wasn’t saving anything.
What actually broke
The tl;dr - My Anthropic account simply didn’t support the model I was pointing to. So the LLM call failed and therefore the app broke.
I built something that worked without really understanding how model versioning and infrastructure ownership worked. And there is no notification of the model deprecation or why it’s broken. I thought about making it future proof, but because I don’t pay for Claude, I can’t, but at least now I’ll know why!
But honestly, I still don’t fully understand everything that happened and am just hoping nothing else breaks.
One thing that helped
It turned out to be much more effective to ask ChatGPT to debug than Bolt itself.
I would paste:
error messages
console logs
sometimes the entire code file
And ChatGPT was surprisingly good at diagnosing the problem, or even telling me how to open the developer console to get the right information. Especially without costing me more tokens/ money!
Key learnings
Moving fast has tradeoffs.
Since I’m not an engineer, I often accept what Bolt tells me without fully understanding how everything fits together. That makes debugging harder later. For now, I’m accepting this because the speed of building is still worth it.
Your app can break even if you change nothing.
Even if your code stays the same, your app depends on “live things” : models, APIs, libraries, browsers. Any of those can update and break something.
ChatGPT is better at debugging.
Bolt is great for building. But ChatGPT is much better at helping figure out what went wrong.
Good Experience
I ranted to my CPO of AI Tech Company husband about it afterward, and he said: “Welcome to the life of tech companies using AI.”
Which I guess means this is… good experience. 😖
