I Tried Vibe Coding…I Hated Every Minute

I kept seeing people posting how their spending their days writing prompt after prompt build software in hours or days, instead of months and years, so I wanted to see what all the fuss is about and try it myself, and spoiler alert, I hated every minute even though I did manage to build something within a couple of hours.

This wasn’t the first time I’ve tried using AI of course, I’ve been using it for several years via ChatGPT and more recently using t3.chat asking questions about the best way to implement something, or get ideas on how a piece of logic might work or using auto complete in the IDE to get simple snippets of code that I can save time writing out myself from scratch, that just need a small review that it does what I am intending.

Before I get into my vibe coding experiment, a bit about me, I love to code, some might say too much, I code as part of my full time job as a Solution Architect and Support Manager and I code in my spare time on my own projects (when I’m not on the Golf course, when they’re not flooded that is) such as Test Track and Solve Flow.

I wanted to try Vibe Coding, but wanted to build something with a purpose, not just create some random app I am never going to look at again. Therefore I decided to make an alternative to Docker Desktop. The reason being, I use Portainer for Docker management on servers and use it to deploy containers and update the environment variables, however, if I do anything on Docker locally its all CLI based and the docker desktop is fairly limited, for example. doesn’t provide any way to configure environment variables so have to back to the CLI and recreate the container, which is a bit of a pain, so I thought I’d create my own Docker Desktop that is more like Portainer (yes I know I could just install Portainer locally, but needed an excuse to make something).

And here it is:

Supports containers, images, volumes, restart policy, environment variables and can pull images directly from Docker hub. Built using OpenCode using GPT 5.2 model and is an Electron app, I asked it to ensure it worked on Mac, Windows and Linux, not actually tested that but definitely works on Mac.

I built it in around 3 or 4 hours, I was impressed with what it achieved, but I was bored, so, so bored. Write a couple of sentences, maybe a paragraph explaining what I wanted, wait 20 minutes while it goes off and does its thing, what do I do in that time, watch the text scroll as it thinks, and reasons, and decides what it should do, or scroll Twitter/X or watch YouTube until the next prompt is needed. Most of the time it did what I wanted sometimes it did things that went off on a tangent, fixed a bug but also made another change that was completely irrelevant, or sometimes I’d have to keep prompting it and asking it to fix a bug that probably wouldn’t have taken me that long to figure out if I’d written the code myself.

So I have an app, seems to work, no idea what the code is like, no idea exactly how it works, what it does under the hood to make it work with Docker. If I hit a bug the AI can probably fix it, but what if it can’t, I have no clue how the code is structured to be able to find where the bug is let alone have an idea on how to fix it. You might say does that matter, if you started a new job, or started working an existing project at a company you’d have the same problem, and yes you are right, sort of, at least if this was a normal project that another person had touched and you’ve started working on it, you have other people to ask if you are not sure on something. Maybe you could ask the AI in this situation and that might be enough.

This isn’t me hating on AI, or even Vibe coding, it has its purpose, sure, but there is no way I’d vibe code an app I was going to release and ask potential customers for money and have to support.

If I was to use Vibe code again in the future, it would be for a quick POC before I write my own version and use the Vibe coded app as a starter, or I would do it to build a marketing site or waitlist site for a new project as my marketing design skills are somewhat lacking.

If Vibe coding is the future and everyone is going to just write prompts and review code generated by an AI, then I am not looking forward to the future, and I think I need to look for a new career, any suggestions?

For now, I am going to continue what I do now, primarily write the code myself (did I say already I love to code) and use the AI as an aid to build a plan, get help for a particularly complicated bit of logic but I will continue to own the responsibility of writing the majority of the code and business logic, have a full understanding of how it works and fits together. Will that make me a slower developer? Sure, but I want to ensure I can provide the best, most reliable product to my users and know how it works and answer any questions if I get asked anything about it.

What are your thoughts on Vibe Coding, have you completely stopped writing code and just prompting, do you do a mixture of both, let me know.