What I learned from vibe coding a small web project with GitHub Copilot and GPT-5 mini

A few weeks ago, I decided to finally try vibe coding, mostly out of curiosity, but also due to necessity and a lack of time. The bet was to create a small web project using a coding assistant, without much manual coding.
First things first, the tools I used for the job are what I'm most familiar with - GitHub Copilot and VS Code's Agent Mode. While I use GPT-4.1 as my daily driver (what a time to be alive, I'm calling an LLM my _daily driver_), I decided to give GPT-5 mini a go. Spoiler alert: both models seem to be on par, so we won't be focusing on that too much.
I'm not going to dive into the project details, but rather share some of the learnings and insights I gathered from this experience. All you need to know is the tech stack, which includes Go on the backend, SQLite for the database, and React with Redux and Sass on the frontend.
I should also mention that I'm familiar with all of these technologies, but I'm really a beginner in Go, so I could use the help in setting up the project structure and gluing things together.
Starting out with the good, the experience was surprisingly smooth. My little AI assistant was delightfully quick in setting up the project structure and installing the dependencies I wanted, without me having to lift a finger. Full disclaimer that I knew most of the toolkit I wanted to use, so it didn't have to guess much in the earlier stages.

I'm not going to dive into the project details, but rather share some of the learnings and insights I gathered from this experience. All you need to know is the tech stack, which includes Go on the backend, SQLite for the database, and React with Redux and Sass on the frontend.
I should also mention that I'm familiar with all of these technologies, but I'm really a beginner in Go, so I could use the help in setting up the project structure and gluing things together.
Starting out with the good, the experience was surprisingly smooth. My little AI assistant was delightfully quick in setting up the project structure and installing the dependencies I wanted, without me having to lift a finger. Full disclaimer that I knew most of the toolkit I wanted to use, so it didn't have to guess much in the earlier stages.
I was pleasantly surprised by how well it handled boilerplate setup and small refactors I suggested. It was also very good at performing basic tests via curl commands and stuff like that to verify that the endpoints were working as expected.
Another thing I noticed is that I didn't need to dive too deep into the documentation of that many tools. As I'm sort of familiar with Go's ecosystem, I managed to find some dependencies that looked right. I didn't want to spend 40 hours debugging a live reloader and writing my database connection call, so I just asked the AI to do it for me. What would have taken me a day or two, took a couple of hours. Bonus points for not having to read through a ton of docs that I wouldn't need ever again during development of the project.
Last but not least, it was truly a learning experience. Seeing how the agent was restricted to terminal commands and the editor, I noticed how it did things I wouldn't have thought of or been used to. For example, I like how it used curl to test endpoints, or lsof to find process IDs, and actually picked up the trick myself after that!

From 30-seconds-of-code: learnings-from-vibe-coding-a-small-web-project