When I first used Lovable, it was the only AI tool that gave me a similar feeling to the first time I used ChatGPT. With very little prompting, I was able to create (almost) fully functioning websites, even though my learning of the basics of HTML and CSS lies far in the past. While I often saw on social media that many people advocated for websites like Lovable, I didn’t know how well they worked.
This entire experience really showed me one thing: the speed, creativity, and accessibility at which generative AI functions today leave great hopes for the future.
While the frontend looked amazing, I have to say that right now it takes more than just one prompt to create and deploy a fully functioning website. If you want to integrate a database or AI, which is probably important for most websites out there, especially if they will be used commercially, you must integrate a backend. But don’t worry, even this is possible with partnerships from Lovable and Supabase, connecting your website to a Postgres database and allowing you to use APIs, Edge functions, and other great options.
With Lovable being what it is, what could be any limitations or points for improvement that I could wish for? Well, while the website helps you build an amazing first draft of what you want, and good prompting makes a great difference in the outcome, the results in the beginning are still open to a little randomness. More transparency on how the results are generated would help to better individualize the website to my preferences. Also, when there are some parts of the website that I don’t like, it would be amazing to have a more accurate and intelligent debugging tool. Often small errors, whose origin I lack the expertise to identify, can be very frustrating to solve if you don’t know what commands to give to the website. Lastly, a better guide on how to develop, integrate, and deploy the website would be an enormous help for anyone new to coding. While I read the Lovable Bible and found online tutorials a huge help, AI has all the requirements to support the coder on the spot, without having to leave the site.
My first time using Lovable was just a fun project for myself, to see how far AI tools have come and what will be possible in the future. Despite the bugs, it is still one of the most impressive tools out there, and I can urge everyone to try it out themselves.
Hi Bennet,
Good insights you show in your article. I have used lovable a few times as well, and must say I agree with your limitations of Lovable.
What I personally do to improve the outcomes and leave less to ‘randomness’ is using AI generated prompts via ChatGPT. I ask ChatGPT to come up with a very clear prompt for Lovable, which helps me achieve better results.
Secondly, using JSON-prompts can also help. These structured prompts are easily readable for AI models, which leaves less open to randomness.
Lastly, it indeed is hard to debug generated projects without knowlegde about code. I think most of these problems will continue to exist in the coming few years, because these issues are inherent. Fun fact: there are a lot of “Lovable code reviewers” on LinkedIn. This new job title is widely expanding. These people check Lovable code, improve security and fix bugs. This can actually make you money!
I conclude that real software developers and programmers will still be necessary in the future, because coming up with a digital solution to a complex problem is a skill in and on itself.
Kind regards,
Olaf