My First Vibe Code Journey: From Frustration to “Boom, It Works!”
I recently started my vibe code journey and decided to roll out my first pet-related test app. The goal was simple: try something fun, learn the ropes, and get it working end-to-end. The reality? It was a mix of trial, error, and a whole lot of patience.
The Tools I Used
- GPT-5 (including GPT-5 Thinking mode)
- Gemini 2.5
Round 1: GPT-5 Thinking
I gave GPT-5 Thinking a full day to help me solve an issue (yes, just one issue) involving html2canvas.
The result?
- At least 20 rounds of “fixes”
- Multiple new chat sessions to avoid previous confusion
- The same error haunting me like a stubborn bug that refuses to leave
It wasn’t just frustrating, it was exhausting.
The Frustration Point
What really broke my patience was that some issues we did fix along the way mysteriously came back after a few attempts.
My guess? Short memory cut-off during the conversation. Once the context was gone, we were back to square one. And yes, that meant repeating explanations, re-sharing code, and losing momentum.
Round 2: Gemini 2.5
At this point, I was done wasting time. I switched to Gemini 2.5, went back and forth three times, and…
💥 Boom. All done. Perfect.
That’s it. Same problem. Fixed in minutes.
Key Takeaways from This Experiment
- The “best” model might not be the best for your specific problem >> each AI has strengths and weaknesses.
- Context persistence matters >> losing history mid-debug is like losing your place in a puzzle.
- Speed over loyalty >> if one tool isn’t working, switch. Your time is valuable.
- Have a backup workflow >> I now keep all debug history and code snippets in a separate doc so I can quickly restart on any platform.
Next Goal: Build, Secure, and Roll Out
Now that the first test app works, I’m setting my sights on the next stage:
- Build a proper database: I’m thinking of using Google Cloud for the backend
- Host it on Cloudflare for speed and security
- Integrate Google Auth to handle user sign-ins
- Once ready, I’ll roll it out for beta testing and start looking for early users to play with it and give feedback
I want this to be more than a test, I want it to be something people can actually use, even in its early form.
Final thought:
This wasn’t just about fixing one bug, it was a reminder that in coding (and life), flexibility beats stubbornness. The “world’s best model” is great… until it isn’t. What matters is getting the result, not who delivers it.