Now that AI is able to generate code, I was thinking about how everyone who is using AI or is vibe coding is thinking about maintenance.

Software inevitably breaks with deprecations, updates, security advisories, etc, now and that with software, apps, web apps and SaaS's being easy to generate.

The technical debt of the software increases quickly, (adding APIs, databases, analytics, features, more libraries, more files, more code paths, etc) so the likelihood of the software breaking also increases, especially with vibe coding being popular.

How is everyone thinking about this?

Are you just prompting ChatGPT, Claude and other LLMs to just "fix the code?" and crossing your fingers? or you adding unit and integration tests or something else?

Thanks!