Today I tested Lovable, a fast-growing Swedish startup that promises to be a “superhuman full-stack developer.” The idea is that you describe your desired app in natural language, and its AI builds it for you.

Experience with Mind Maps

I experimented by asking Lovable to “vibe code” a mind-mapping application. The tool succeeded in generating a basic mind map interface that runs entirely in the browser. However, when it came to building a backend for storing these mind maps for future expansion and for other users, things became more complicated. The system wanted me to work with Supabase, an open source backend platform.

Backend Complexity

While the “superhuman developer” aspect was appealing, it turned out that Lovable only partially aids the process. The promise of a no-code solution quickly fades when you realize that a study of technical documentation may be necessary to bridge the gap.

Security Concerns

Adding to the complexity, a person claiming to work for Palantir recently mentioned on X (formerly Twitter) that he discovered several security issues in production appscreated by vibe coders using Lovable. Upon reviewing the code the AI generated for me, I noted multiple “moderate” vulnerabilities (think cross-site scripting, XSS). This underscores that we are not yet at a stage where one can rely entirely on an AI to produce secure, production-ready software.

More about security, debugging and ai-agents can be found at Gary Marcus’ blog.

Conclusion

Lovable is an intriguing platform for rapid prototyping, but be cautious: always have a professional review any code you plan to deploy in a production environment. The dream of a fully autonomous, error-free AI developer remains distant—we are far from the era of AGI, let alone superintelligence.