Are AI-generated features causing more problems with MVP maintenance than ever before?

AI has allowed me to quickly create an MVP by automating the process of coding and adding features has been a breeze.

However, the trouble appeared only much later.

Each newly added feature would introduce a new dependency or edge case, or even a duplicate of the existing logic. The project looked slick but changing anything meant breaking something else.

This experience taught me that fast development is not only about the speed at which we can introduce features but also about the simplicity of further changes to the code base.

When do you know when to stop adding features and start refactoring the core?