You can’t look at much on the Internet without hearing about vibe coding. A lot of it is negative… folks using AI to create ‘slop’ that they don’t understand or that can’t stand up to formal scrutiny. Terrible for long-standing corporate code but it opens up a whole new world for those who have an idea and want to make something of their own.
I’m not here to say that vibe coding is the best thing in the world – It can definitely churn out some crap. But I did want to share my experience with vibe coding coming from someone with a technical, but not coding, background. What I’ve learned and some general thoughts.
I’ve been a tech nerd my whole life! Ever since I can remember I was tinkering with computers and systems, trying to figure out how they work and how they can work for me. I dabbled in coding here and there but never pursued it. I did do a good amount of scripting work for myself and as a junior sysadmin back in high school. In my professional career I’ve focused on the SDLC (software development lifecycle) as a business analyst and a project manager. Understanding the full development cycle and working closely with devs and QA folks taught me what questions to ask and how to test an application.
I’d consider myself a power user of AI and LLMs. I use them on a daily basis for both work and personal tasks. I’d say I understand more than the general consumer but nowhere near what an expert does. I’ve also studied a bit about how they actually work which has been interesting as well.
Call it FOMO? Kinda, but not really. I kept reading articles about how great LLMs were getting at creating code and Claude Code and OpenAI’s Codex were being touted as next-level. I sat and thought about it and figured, ‘why not dabble’ in these new technologies? I like to understand what’s out there and how things work. I also thought about what would be ‘nice to have’ apps for myself that might not exist or that have ads and used that as a base moving forward.
Some people are ride-or-dies for particular LLMs…I am not that person. I do have a preference for Claude’s personality, but I use image generation enough where I needed ChatGPT (side note: not a fan of Gemini overall – we just don’t vibe).
For this experiment, I signed up for a month of each – their $20 tiers to get my toes wet. I figured that I could have each LLM help me develop one app each and see how the experience differs.
This was less of a choice for me. I wasn’t going to feign being a software developer so I opted to use the Claude Code and Codex GUIs for this task. There’s (probably) obvious overhead, but I ultimately consider it negligible. If I wanted to write a post about amateur vibe coding, I figured I should use the more basic tools.
In order to make something, I needed an idea. That’s arguably the hardest part! And in this case I didn’t want to compare how Claude and Codex would make the same app – just experience how each one works.
So I came up with two ideas: a scratchpad app and an app to track a hobby of mine. The latter was meant to track detailed information online reality games including the cast of the game, advantages, and rounds. That app is definitely more elaborate and has evolved a bit since I had the idea.
To begin I used the planning mode in both apps. I decided that I would let it create a baseline and modify it as I went. It starts with a prompt, something like:
I’m looking to create an iOS app that I can use as a personal scratch pad app. I want the app to load extremely quickly and have basic formatting controls.
I’m sure my OG prompt was a bit more in depth, but that can kick off the planning process. The planners in each review the prompt, and then ask questions to create a requirements document. They work and think for a bit and eventually produce the artifact! Easy peasy.
I had the option to review the documents but at this point I wanted to just keep moving forward. Both ask me if I wanted to implement the plan, and I click yes.
You know, I wish I documented this process rather than recounting it, so apologies for no screenshots!
But the first drafts that came out were pretty good! I prefer to work from a baseline rather than create from scratch. That’s a me thing though – I’m not a designer, I’m a refiner.
The initial apps were actually pretty good though. The functionality was there even though the look definitely needed refining. The defaults were all done in Apple design language with Liquid Glass and Apple-like menus.
Definitely impressed with the set it and forget it – and was excited to see where I could take these apps
I then focused on getting the features I wanted in there. I found it was very easy to go through the app like a QA tester and jot down the things I wanted. Told you my experience was relevant 🤣
I’d go through screens and write out flows and desired effects. For example:
This worked pretty well and became my go-to for features. I typically could get everything I specified with my rate limits, but I would have to break them up and wait for my rate limits to reset.
Sometimes it was frustrating to get the functionality to work as expected where I would waste tokens and have to tell it that the fix wasn’t working. Classic LLM. That being said, you gotta have patience when vibe coding.
The vibes gotta be vibin’.
They are SO FRUSTRATING!! And literally feel like they’re intentionally addictive. They probably are. I felt myself constantly looking at the clock or staying up 15 min later to get a prompt to run while I was asleep.
I may have switched to a $100 plan to overcome that but who’s to say (…I did). I’ve never hit a limit on the upgraded plan on Codex and have done PLENTY.
This is getting long and you’re getting the idea. This adventure into vibe coding has been so FUN! It’s brought back a piece of passion for technology that may have went away a long time ago. It’s fun to just see what happens while still being able to learn more about app development at the same time.
It’s about what you make it out to be. My expectations weren’t to learn to code. It was to make an app that could help me. I believe AI and LLMs are here to augment and assist us and expand what we are able to do. Apps-on-demand is what the future is going to be for both personal and corporate use.
While I’ve been iterating on these products, I’m actually starting to think that these could be viable for the niches they serve. For my scratchpad app, I wanted something to type in that didn’t take up space in my notes apps (Notes or Bear) but has some quality of life features that are useful. Who knows – I may publish it in the App Store soon!
Honestly, they both were very similar to me. For someone who is not working with code day to day I’d recommend trying the free versions of each and seeing which UI you like better! I ended up going with Codex because of their usage incentives, but I also like the app as well. I like that it’s separate from the main chat portion of the LLM.
There are so many reasons that vibe coded apps are hated. Those reasons are obvious when care isn’t taken in making them resulting in slop and garbage. On the other hand, if you utilize your LLM and some research you can learn how to manage your app effectively, produce test cases, and push a good product.
I’m far from done in my vibe coding learning journey. I’ve touched the tip of the iceberg and still managed to produce pretty good results! Excited to see what the future has in store for this stuff.
