Riding the vibe-coding hype for a prank

I was looking for an excuse to actually try out one of those hyper-growing AI tools when one opportunity revealed itself: pulling a good ol' prank on my past employer. They were in a fundraising stage and apparently, a big competitor was arriving on their turf... 😏

Summary

Preface

Before diving in, a couple of notes:

  • I feel like I did not leave Recovr, the prank’s target, but rather decided to join We Invest for the job it offered me instead. I loved my time at Recovr, I’m proud of what I’ve accomplished over there and stayed close to the entire team
  • One of my co-conspirator is actually in a Lead position over there 😏
  • I was 100% sure that this would be well received. The goal was not to hurt anyone nor cause any kind of distress. We have a (small) history of trying to prank Aless, Recovr’s CEO
  • I was looking for an excuses to actually use any of those AI-prompt-to-code tools such as Bolt.new, Lovable and such for some weeks

The Genesis

Aless, Recovr’s CEO, has an history of teasing new stuff and driving people crazy over it.

You’ll never guess what happened!!! – Aless, probably

By mid-2024, Recovr realized that, going into their next phase as a company, with a fundraising going strong, they would need to revamp their brand. This created an opportunity for Aless to drive the Recovr team crazy by dropping small details about new logo, color scheme and such, without ever revealing anything meaningful.

Entering our insider, Lead Frontend, who would at some point be involved for the color scheme. Anything logo-wise would not be shared with him, but we had an in.

So naturally, at a diner with him and another good friend and past colleague from Drawbotics, we schemed the whole thing in less time than you can say “can I have another beer with that steak”. Leo found an awesome name for the new company: Retreev. Arguably a better one than Recovr for their business, but oh well 🤓. We offered it to Aless, in case he’d like to change 😏, for a small fee, of course.

We would get the color & fonts, “steal” those to create a fake competitor with similar brand that would have raised money earlier, with a bigger amount and launching in the same markets.

This would be a one-stone-two-birds opportunity: pranking Aless & Recovr and trying out the AI hype around vibe coding a website in a few hours.

after_all

Prompting a new website

To be honest, today I might go with Lovable for its EU roots but I’m happy to have tried out Bolt.new for the purpose of this experimentation.

I was (and still am, to some extent – more on that later) skeptical about the AI general hype. So naturally I dove in and kept an open-mind. I initially thought that I might be surprised, but I did not expect to be surprised so fast.

I set aside a saturday afternoon to build the website, from start to finish. I found and used, during this timeframe, those tools:

For a total cost of €6, the domain name, plus 1 month subscription to Bolt.new.

Now that I reflect on the process, I might have spent more time designing the logo, illustrations and finding the tools to do it.

The good

The first time I prompted something and saw that it was actually not a chat but an agentic approach, refactoring the filesystem as we went over new features, actually “thinking” through new iterations, I was speechless.

ChatGPT was some kind of revolution but I did not buy the fact that it would transform the way I work in late 2024. Bolt.new, on the other hand… I was hooked.

The first few tests made me subscribe immediately and I tried out some ideas. After a few examples, I switched to build the competitor’s website.

I had the first 60% of the end-result in less than an hour I believe. In these 60%, there is the actual content of the website. That’s something that LLMs are phenomenal at: synthesizing and re-formulating based on other sources (such as… Recovr’s website 🤓). To fill out a marketing landing page, it’s extremely convincing.

Then I started to hit a wall.

The bad

I started to hit a wall the moment we dove into details. Margins, paddings and such. After a couple more tries, I realized that I changed my prompts from “naive un-technical” to extremely precise. To get to the expected result, I had to start using more precise terms and technical slang. I felt like I hit the limit of such tool that promise the ability to build a website to anyone when I had to specify paddings and margins. The difference is obvious for anyone who had to any CSS, but these tools are targeted at people that never touched an IDE ever.

I feel like the promise, once again circa end-2024, was not met for non-coders. It was not obvious to “polish” the website it gave after the initial few prompts.

Funnily enough, the time allocation was very similar than when working with a human:

  • 20% of the time for the first 80% of the product
  • 80% of the time for the last 20% of the product, the polish

Closing thoughts

All-in-all, the process was a real eye-opening one:

  • The promise to build something in hours rather than in days holds really well
  • The promise for anyone to be able to build something is quite exaggerated. One can dive into technicalities and actually do it, but they’ll have to dig and learn when closing in on details I feel

Those tools are really impressive and seeing the advancements in the engines that power them (lately with GPT-5 for example), we can only imagine that those last 20% where you need to be precise & technical will become easier and easier, making the whole process more acceptable for everyone.

If you’re curious, here’s the end-result: https://retreev.eu/. I could have polished it a bit further, every time I stumble upon it, I see huge issues, but I wanted to stay within my “one-saturday-afternoon” constraint so I don’t fix anything!

The reveal is on the team page :)

The website was presented during the Recovr’s christmas party along those lines:

Hey Aless, I heard about these guys, I don’t know if you know them

I was told by people present at this party that there were a few drops of sweat. The pranked person told me that they did not believe it for more than a few seconds.

As usual, the truth lies in-between 🤓

PS: here’s a screenshot after_all

Stan
Stan