I vibe-coded an app with AI and was mind blown 🤯

For the uninitiated, vibe coding is the art of building software apps without writing any actual code yourself. You prompt the AI, check the vibe of what it produces, and keep prompting until the vibe matches what you have in your mind.

I’ve been exploring AI for the last month or so and finally decided to code a basic project to see how good it actually is.

I was mind-blown by what I achieved with just 10-12 hours of work on the weekend. This is what I got done and its implications for the future of work.

Note: I started from scratch without prior experience in writing Javascript, hosting applications on the web, or using the command line extensively.

What I accomplished 👽

The very first project I built was a basic alien shooter game:

The initial version I made was elementary; it looked like an MS Paint drawing:

All this was done with just a few prompts in this AI coding tool called Cursor. It allows you to type what you want and turns it into code.

I told Cursor to make a shooting game with triangle spaceships, make aliens randomly appear (with a max count of 10 at a time), and told it to build logic such that each alien takes two bullet hits to disappear.

Cursor took care of the rest and gave me a basic game with sound logic.

It automatically assigned different (intuitive) keyboard keys to control the spaceships, and made the aliens descend without any prior prompting. It understood the “point” of the game without explicit prompting.

This took no more than 15-20 minutes of actual prompts to build and finalize.

Then, I took this and told it to “put it in a format that looks like a mid-2000's console game.”

It then produced this:

It also helped me set up a local server, which allowed me to run the game in my browser. It helped me install the necessary packages on my device (such as npm and python3)

I have no idea how the code works or how it did all of this. It just did it!

That’s fascinating. But it's OK, nothing impressive. It's "cute", not practical.

Levelling up from rookie games…

How about we make it do something moderately useful?

After another 30-45 minutes of prompting, I was able to build an “Influencer Analyzer” that’s able to show you the influencer’s metrics upon querying:

I gave it a simple drawing of how it should look on the front end and made some adjustments to how it looked. The AI took care of the rest.

I faced some errors, which I was able to debug using plain English prompting:

I did all this in the first few hours of playing with the Cursor IDE.

Note that I have ZERO prior experience with this stuff, and I was able to accomplish something in 45 minutes that would take a typical coder (without AI support) 1-2 days.

What would have taken an actual craftsman 10-12 hours took me just 45 minutes. It’s a 10x productivity improvement for someone without any expertise.

Think of it like this:

I was able to bake a gourmet cake in 45 minutes, while it would take a professional pastry chef 12 hours to make the exact cake from scratch— measuring ingredients, mixing, baking, decorating, and perfecting every detail. The difference?

I had an "AI baking machine" that did all the work for me. All I had to do was describe the cake.

I didn't stop here. I went one step further to make something even better.

I built an app called "Agensigner" that simulates design mockups:

I then used AI’s help to help me host it on an actual website so you can try using it:

Agensigner - Idea to design in seconds

agensigner

See how my projects are getting progressively more “real”?

All this in less than 12 hours of work over a weekend of tinkering, with no prior experience. Now imagine the possibilities if someone goes all-in on this stuff.

You could literally fast-track your way to building a solid web app with authentication, databases, and an actual use case in just a few days.

Projects like these take weeks for dev teams to build. Now, just a few prompting hours away.

It’s insane.

Does this mean engineers are obsolete?

I disagree that engineering would become obsolete. Coding might be; engineering - not so much.

Code is just the language of computers. 50 years ago, computers were programmed using machine language, which looked like this:

Nearly 100% of engineers today will not understand this. Why?

Because they write code in much more human-friendly “high-level” languages such as Python and Javascript:

The next natural evolution is just coding in plain English. That is what I did.

More straightforward coding doesn’t mean engineers become obsolete; it means coding becomes much simpler and more democratized.

What the iPhone did to photography 📸

A straightforward example of similar democratization is when the iPhone was launched.

It enabled anybody to be a photographer or a cinematographer. Anyone could make and upload a video online without requiring a complex, expensive camera, equipment, and lighting. Anybody could click photos or record a video.

That is what AI would do to coding - anybody could build an app they want or create a quick software tool for their school. AI will democratize coding so that anybody can build software.

“Building an app” will be as simple as “recording a video”

The iPhone did not render good storytelling or cinematography obsolete; it only made recording videos 100x easier than what it was 30 years ago.

Similarly, sound engineering and problem-solving will always be valuable— coding, not so much.

A note for junior developers…

Tools like Cursor render junior developers obsolete. If you gave Cursor a detailed product specification, it could code it faster and better than an entry-level engineer.

So, if you’re a young developer reading this, embrace AI to write code as fast as possible and train your mind to solve higher-level complex problems.

If your edge in the market is knowing a complicated coding language and web terminology, you will be rendered obsolete. Focus on engineering, not programming.

If you are not using AI to code, you are handicapping yourself for no reason other than being an arrogant human who thinks you’re better than LLMs.

What it still can’t do 🥲

Much of my time was spent on logistics, like figuring out which package to install or which library to use or debugging basic things like git conflicts (stuff that is very obvious for engineers, not for me).

I think AI still has a short way to go to help complete rookies like me build production-ready apps without going through all the logistics that come with it.

AI still cannot fully understand systems and architecture. It can’t build sophisticated technical architecture or read complex documentation to solve tricky problems. It doesn’t understand codebases.

I can’t unleash an AI on SARAL’s codebase and expect it to build new features for me.

But the pace of change is so fast that it might be able to do this in 2-3 years. For context, ChatGPT is only about 2 years old, and it can already do a lot.

It went from being a "search engine" to being a brilliant co-worker in just 24 months. It can be so much more in the next two years. I don't think most people anticipate the pace of change. It will be rapid and feel sudden.

The thing with AI is everything is a “yet.”

It can’t do XYZ… yet. I can’t unleash an AI on SARAL’s codebase and expect it to build new features for me... yet.

It will be able to do so with more time and better models. This has profound implications for the human race and economy; I will write a separate blog on this topic.

Why I spent time doing this 🤔

I have no intentions of becoming a programmer. So why did I spend a weekend creating apps?

As the CEO of a tech company, it is my job to know what’s out there and develop intuition about leveraging AI inside our product. One of my primary jobs at SARAL is that of a product leader. I speak with customers, understand their JTBD, generate product ideas, and share them with the engineering team to build.

Being exposed to this technology has opened my mind to SO MANY more things we could be doing:

1. At a meta-level, improving our engineering team’s output.
2. At a product level, improving SARAL for customers and making their workflows 3-5x faster.

I will spend a few more weekends tinkering and seeing what I can build with AI. Then, I will take this to my team and see how we can improve our product with it.

I am SUPER PUMPED 💪

Where you should go from here

I will write another blog post on AI's implications for the world and the economy. For now, I encourage you to try tinkering with AI in your line of work.

If you’re a marketer, a designer, or a developer, try to build something using Cursor for your own workflows and see the possibilities. It might blow your mind like it did mine.

Here's to vibe coding! ☕