Lovable.dev: Remaking my startup with AI
Is software business dead?
I have been experimenting for a while now with AI app creators. The 3 platforms I tried are:
Lovable.dev
Bolt.new
Heyboss.xyz (after the drama with Lovable)
You know the argument: Software as a service (SaaS) will be commoditised.
Everyone will be able to build their own app.
Prices will floor and noone will be able to make money from a software startup.
We will put this in test today. I will take a personal risk and try to recreate the product of my startup, GodmodeHQ, using Lovable.dev.
Watch this.
Here is the latest state of the app as published by Lovable. Feel free to check it out here. This is the latest state of the application as best as I could bring it.
Here is the my starter prompt:
The first version Lovable creates is the following:
It gets the menu items right however the process is still too fragmented and harder to use than what we designed. So I will prompt it to fix this.
Can you make it a sequential step by step process? First let the user fill out filters for selecting ideal customer profile. Then show them a test lead, if it looks good to the user by his or her confirmation, go to the research step and do a through research on the lead and the company. Then if the research is approved, use the research to generate an outreach message.
This prompt results into the following modified app. It does follow my instruction as to putting the user experience into a more sequential process.
For simplicity, I will try to integrate this with Apollo.io’s API as it has a simple developer experience to fetch the leads.
I connected Supabase for it to be able to store data and use a backend service to call the Apollo API.
This is when I hit the first error:
From hereon it was a steep challenge to get it to working.
I followed the instructions in the chat and let it implement the lead fetching using Apollo.io docs. However it was not able to successfully implement the system to fetch the leads and save them into the database.
The most frustrating part of this is where the chat agent will point to its mistake:
and recommend refactor the index.tsx file. However upon refactoring it, it will run into the same error and not recognise that it needs to do something different.
I have come across the same behaviour with Claude 3.5 Sonnet which I suspect both Lovable and Bolt fully leverage under the hood.
This is somewhat promising but seems like it is very far away from building Godmode on its own and commoditising our whole business!
We are safe for now!