From Budibase to Supabase and Next.js

Predator started with a low-code attempt in Budibase. The reasoning was to avoid the vendor lock-in of fully custom software, with the vision that users could eventually build their own tools. It was also useful in the beginning because it made the shape of the work visible quickly.

That vision did not hold up. Even with AI, users cannot know what they need without a basic understanding of processes and data structures. Building systems is still hard for users, even today.

The other limits showed up once the product became more than CRUD screens. Custom study planning needs performance, special workflows, document generation, permissions, and state that has to remain correct across many users. At that point the low-code layer started to be part of the problem.

In the age of AI, low-code is less attractive for this kind of system. It still gives you a managed platform, built-in screens, and some safety around common operations. But if most of the hard work is custom behavior, AI makes real code cheap enough that the performance and escape-hatch costs matter more.