Vibe coding: the new no-code?

At MarsBased, we have been experimenting with vibe coding for a good while for side-projects to assess the validity of this new trend. Is it just a passing fad or is it here to stay?

I can summarise my thoughts in the good, the bad and the ugly. Shall we start?

The good

Vibe coding, much like no-code back in the day, has managed to attract new people to the world of development. More and more people are trying to build their ideas using Lovable, V0, Replit and the like, and these people range from complete outsiders to the world of tech to former developers willing to give it a try to see if this brings them back to the joy of developing.

Even though the former won't quite succeed, a few did. But we can't say that everyone is now able to code (and much less, code right). There's a subset of non-technical people who are able to figure it out because they're sharp enough, have a friend who can help them with this, or have an infinite amount of time, so they're able to iterate time and again (paying big sums of money for the credits) until they get it right.

On the other hand, former developers, are coming back to coding if only to just give it a spin.

For instance, I haven't coded seriously for over 15 years now. My current CEO title forces me to do sales and strategy the whole day, leaving little to no time for coding besides fixing bugs on the MarsBased website or contributing to this one. I did, however, carve out time during the summer to do a few apps with vibe coding, using Lovable.

While no-code didn't really bring retired coders back to active coding, vibe coding might actually do it. Even if it's only to say "I could do it better than an AI".

We will undoubtedly benefit from having more people coding and trying things. Some of them will work out and will create jobs. Others will turn into scaleups and, who knows, we might see a big exit or two one day.

At what price? Keep on reading.

The bad

AI is enabling us to go faster... but where to? Unfortunately, faster usually doesn't mean better. On the contrary: faster tends to correlate with poorer quality.

While a lot of people will think "well, this will propel me to success much faster", the reality is that most vibe-coded startups will crash and burn, leaving behind a lot of AI rubble scattered around the internet.

The reality is that only trained eyes can actually make good use of AI-generated code. If you know the programming language well enough, you will not accept just any suggestion made by Cursor. But even then, you might just want to go fast and accept All The Things™ and end up accepting unnecessary code, or downright bad code.

This code will increase technical debt at the speed of light, which will make everyone unhappy and the cost of working on this code henceforth much more expensive and cumbersome. But we will not get technical in this article.

AI sludge (or AI slop) is already polluting the internet. We don't need more code.

To follow with the parallelism with no-code, many successful no-code startups tend to combine no-code with traditional code or transition away after early rounds of funding. Further, less than 1% of no-code platform users reach significant revenue, and the majority of no-code apps struggle to scale efficiently due to limitations in control, customisation, and scalability.

No-code apps were seen as red flags by investors too. While there are successful examples of no-code apps reaching substantial scale and securing funding, they constitute a very small fraction of users. The same will be said about vibe-coded apps.

If you're using AI to write more code, not better code, you're doing it wrong.

The ugly

By attracting more people to code, more people will be able to create apps and companies. In other words, the barrier to create new businesses has been lowered and more widely democratised.

However, in my view, the 0 to 1 hasn't been solved. Vibe coding done by non-technical folks will get stuck at prototype stage. Some will be able to raise funds with a semi-functional app and even some free or paid clients, but won't be able to scale without rewriting the app in full.

We were promised that everyone could launch an app, but they didn't tell us that this app wouldn't work or that it would be full of security liabilities, like in the case of Tea. This app, designed for women to share reviews and safety information about men, experienced a catastrophic breach exposing over 72,000 images, including selfies and government ID photos. Additionally, private direct messages discussing intimate topics were leaked. The breach occurred because Tea's backend was left unsecured partly due to the use of vibe coding.

The leaked data spread widely on platforms like 4chan and included verification selfies, IDs, and private messages. The app had boasted features like no screenshots allowed and data deletion policies, but the breach revealed large amounts of sensitive information were inadequately protected.

While there is a case to be made that AI will contribute to fight gatekeeping in a very opaque industry, AI can also cause big accidents that could have been avoided if a professional had a look at the code in due time.

Finally, we have been requested over the years to rewrite no-code apps into fully-fledged apps that scale, and we are seeing this right now with AI.

We have received a few project requests to review vibe-coded apps and the reality is very crude: the code is convoluted, full of overengineering, exposing credentials right, left and centre, and just not engineered to scale.

If you want to do this right, let's talk.