Before You Build in Lovable, Read This
What you'll learn:
How to stop burning credits on brainstorming
Why you should plan in Claude before you touch Lovable
What the security record means for how you use the tool
Before Connected Women in AI’s 100 Women in AI celebration yesterday, there was a workshop from Lovable.
Lovable is a tool that lets you build a real working app or website by just describing what you want in plain English. No coding required. You type "build me a quiz that tells people their AI personality" and it builds it. That's it.
It sits in a category called vibe coding - AI tools that turn ideas into working software without you writing a single line of code. The pitch is that anyone can be a builder now. And largely, it works.
Here are three key pieces of information about using Lovable.
1.Use “Plan” Mode
I have built in Lovable before. But I made the mistake of using it like ChatGPT and Claude - having a conversation to build up my idea. I immediately burned through all my credits brainstorming in the tool before I built a single thing.
The most valuable learning from the workshop was that they now have “Plan” mode. So use “Plan” mode before you use “Build” mode because planning doesn’t use any credits!
2.What I Did
I built out my entire concept in Claude first. Why? Because I am in a deeply committed relationship with Claude and it already knows all about me. Plus I already had a Project going for what I wanted to build. It was a short-cut rather than starting from scratch in Lovable. Then I exported a planning brief and dropped it into Lovable in plan mode. Not build mode. I stayed in plan until I was genuinely ready to build.
Claude for thinking. Lovable for building.
In thirty minutes, I built a content recommendation tool for Connected Women in AI. Thirteen questions gives you AI archetype and recommends three learning path resources for your skill level.
Try it here: What's your AI personality? (at the end it asks for your email which currently goes no where).
Don’t worry the legal professional version is coming! 🪩
3.There are Security Concerns…
Now the two things you also need to know about using Lovable.
1. Their security record is not clean.
Lovable has had real incidents. An API flaw made source code, database credentials, and chat histories from projects created before November 2025 readable by any free account. Five API calls. No hacking required. One of the projects caught in that breach belonged to Connected Women in AI the community I had just been celebrating with.
When it was reported, Lovable called it intentional behavior. Then blamed their docs. Then blamed their bug bounty partner. Their official position is that security is the user's responsibility. Which would be fine if their users were security engineers. They are not. That is the whole point of the product.
2. You do not own what you build there.
Every project I have built with Claude is backed by a database I control. I can see the logic. I can see the credentials. I know what is happening and where. With Lovable, that visibility disappears. The tool abstracts it away, which is part of the appeal until something goes wrong and you realize you never understood what was running underneath.
That lack of transparency is not just a security issue. It is a control issue. And for legal professionals especially, control is not optional.
Use Lovable for throwaway projects only. Nothing with real user data, work systems, credentials, or anything connected to clients or employees. The quiz I built is the right scope. Fast, fun, no sensitive data behind it.
The filter I now use: would I be embarrassed if everything I typed into this tool were readable by a stranger with a free account?
If yes, build it somewhere else.
Speak soon, Laura
