AI-generated Visual Studio SaaS app has many bugs — keep debugging or revert to earlier working prototype and rebuild incrementally?

I’m building a SaaS app in Visual Studio with AI-assisted coding.

My first attempt added many SaaS features at once. The result now has many bugs, and progress fixing them is very slow.

Would experienced developers recommend continuing to debug the current codebase, or reverting to the last stable pre-SaaS app prototype and rebuilding carefully, adding one feature at a time?

My current workflow is:

  1. Use an AI model to analyze likely bug causes.
  2. Use another AI coding pass to modify the Visual Studio project.
  3. Test manually and repeat.

The concern is that the codebase may now be too tangled, and incremental rebuild may be faster and safer.

What workflow would you recommend for recovering an AI-generated SaaS app with many regressions?