My Successes and Failures at Vibe Coding

Just to get it out there—I’m not a developer. It always feels a little awkward and uncomfortable to say that out loud given I’ve worked in and around technology my entire career. For most of my career, I’ve obsessed more around the “why” technology trends should or shouldn’t be embraced versus the tactical “how” to accomplish the task with hands on keyboard. I’ve had a natural curiosity to understand the principles behind why so many technology breakthroughs lead to innovation (e.g. The power of well crafted APIs, Cloud Computing, Data transformations, Generative AI). I’ve found it intellectually engaging and extremely helpful in setting good business and technical strategy. That said, my career path always pushed me toward broader technology strategy and product development and I’ve often craved more time to explore my hands on keyboard coding skills.

One of my latest intellectual obsessions is around Vibe Coding and how generative AI can do for coding what it has done to help every college student’s ability to overcome writer's block. I’ve spent a lot of time reading and listening to the principles behind Vibe Coding and how it works.

One morning I woke up and decided to take the plunge and see if I could shift from understanding all the “why’s” behind the potential power vibe coding and instead see if I could actually build something myself.

So, I decided to see how far I could get in building a full-stack app without being a seasoned coder. Here’s my 48-hour journey.

Getting Started

I downloaded the free version of Cursor onto my Mac. Why? Because I heard on a podcast that it’s the real deal. I intentionally only used the chat prompting function. I didn’t watch any tutorial videos or attempt to learn anything before getting started. I didn’t want to learn to be a coder, I wanted to learn if I could build something without knowing how to code. So, I just started typing out my ideas into the chat window.

So what did I attempt to build?

I’m a musician in my spare time, and I had an idea for a music booking application that would match musicians with venues. Think OpenTable, but for live music. I know it is a niche idea, but trust me—one day, it will exist, and everyone will use it.

I started by entering a one-paragraph description of my idea, referencing its similarity to OpenTable. I was shocked at how quickly the AI understood my vision—and even praised me for the idea. Yay, AI product validation!

Then, it spit out a development plan. I understood just enough about technical architecture to grasp some of what it was saying:

• Frontend: React

• Middleware & Backend: ??? (Honestly, I didn’t fully read it)

• Database: Postgres

My thought process:

“Wow, I sort of understand what it plans to do. Let’s say yes and keep moving!”

It felt a bit like when a mechanic explains what’s wrong with your car—you nod along, pretending to understand, and then say, “Sounds great, go ahead!”

Building the App

Next, the AI guided me through installing all the required software onto my laptop. I won’t bore you with the technical details, but I followed its instructions—sometimes clicking “yes” when it asked to take actions on my behalf.

At times, I had this unsettling feeling—

• Am I downloading malware?

• Am I unintentionally granting permissions I’ll regret?

Either way, everything seemed to work. The AI inspected my setup and confirmed it had what it needed. It asked if I was ready to start.

I said yes.

Then, I watched it work.

Lines of code flashed across my screen, giving me a tangible sense of progress. Every so often, the AI checked in, informing me what it was doing (even if I didn’t fully understand). Occasionally, it asked for permission to continue.

And then—it stopped.

The “Holy Sh*t” Moment

I asked, “What do I do now?”

The AI responded, “Would you like to run your application?”

I said yes. It compiled the code and gave me a local URL to paste into my browser.

And then—the moment of shock.

A beautiful website appeared, complete with:

✅ Intuitive design

✅ User registration & login

✅ Account management

Some features actually worked. I couldn’t believe it. In just hours, I had a semi-functioning web app. My mind raced with possibilities.

Could this be the beginning of turning my ideas into reality?

And Then… The Doom Loop

I noticed a small issue—one piece of content I wanted changed.

Seemed simple enough. I asked the AI to modify words X to words Y in part of the global navigation of the site.

It obliged.

And then everything broke.

• Icons resized randomly

• Content stacked vertically instead of horizontally

• My beautiful modern site now looked like it was built in 1996

I stuck to my guns and only used the chat function as I tried to describe the problem. The AI attempted to fix it. It only got worse.

I kept prompting, trying different ways to explain the issue. The AI kept tweaking the experience. Each change made things even more broken.

It felt like a doom loop—no escape, no solution.

And then—I hit my free trial limit.

Game over. #Fail

So Close, Yet So Far

This experience gave me a glimpse into the future—and it’s so close to being real.

Right now, AI is already a game-changer for real software engineers. But for non-developers like me? I could almost touch the future… but it was just outside my grasp. Introduction of bugs, security vulnerabilities, and maintenance nightmares are all problems of the present. I’m eager to see how these problems get addressed.

I can’t wait to try again in a few months.

If you’re a non-developer, I’d love to hear about your experiences—successes or failures. If you are a software engineer, you still have my utmost respect.