Problems with Lovable applications

First of all, Lovable—the AI-powered full-stack app builder—is amazing! If you had told me five years ago that complex working applications could be built with simple text prompts then I would have stopped listening. Saying that though, I still wouldn't use Lovable for a production application. Let me try and explain why.

I am a purposely slow adopter of AI for coding. My years of web development have taught me that if I do things the right way, slow as they may be, then I will avoid complications in the future. If I take shortcuts and don't think hard about things then issues are likely to surface someday. If I ignore these issues and keep building on top of a shaky foundation then the project is inevitably going to need a total rebuild. This is how I look at Lovable too.

AI agents, which I do occasionally use, are mind-blowingly good at finding ways to solve problems and add features autonomously. They spew out lines of code quickly and effortlessly; code that would take a more traditional developer much longer to write, but as long as it works then who cares right? Wrong!

When a vibe-coder, or a Lovable user, interacts with AI they usually stop when they are happy with what they see on screen, and commit the change to the project. They don't look at the underlying code because, like most people, they don't understand it. This code could be doing something that they cannot see. What if data is silently being erased or modified? What if payments aren't quite the correct amounts or frequency and nobody notices? What if you have thousands of users being notified via email about the wrong things? AI-generated code also has a reputation for generating security vulnerabilities and introducing scalability limitations. Sometimes these issues will become apparent very quickly, sometimes they will go unnoticed for years.

Now let's say that you have been building your application on Lovable for years, and it has been spewing out reams of code all the while. When things went wrong you got someone with more experience to patch it up and it worked again. Things didn't always run perfectly but if you were lucky then it worked well enough to get your application out there. If you were unlucky then you never made it to launch day.

If you were one of the lucky ones then you may have built up a solid user base of a few hundred or a few thousands users, and now something goes really wrong. The Lovable experts can no longer fix your application, or when they do something else breaks. Important data has totally vanished. The code is spaghetti and there are thousands of lines that actually do nothing, but no human could possibly comprehend it. The AI can't comprehend it either and it never really could; it was just mimicking the code it had been trained on.

This is a typical "old-school developer" premonition of doom for vibe-coded and Lovable applications. Maybe someday AI will prove us wrong but I still can't see how that would be possible. Perhaps I need to submit myself fully to AI and vibe-code my next project to truly understand, but for the moment I will remain skeptical. My skepticism has been validated multiple times this year when I was contacted by people who used Lovable to build amazing looking applications — applications that keep breaking. All I could say to these people is that we should rebuild from scratch using tried and tested tools and good development practice.

I still use AI but I always keep one well-seasoned eye on it.