When AI-Built Apps Need a Rewrite vs When They Are Good Enough
When AI-Built Apps Need a Rewrite vs When They Are Good Enough
Someone builds an app with Claude or ChatGPT, it works, and the first comment is always "you will need to rewrite this properly." But will you? The answer is more nuanced than engineers typically admit.
When It Is Good Enough
If your app serves a small audience, handles simple CRUD operations, and does not deal with sensitive data, the AI-generated code is probably fine. Internal tools, personal projects, MVPs for user testing - these do not need enterprise architecture. They need to work.
AI-generated code tends to be straightforward and readable. It follows common patterns because those patterns dominate the training data. For simple applications, this is actually an advantage. The code is conventional, which makes it maintainable.
When You Need a Rewrite
Three signals suggest a rewrite is coming:
Security concerns. If the app handles user authentication, payment processing, or personal data, AI-generated code needs a security review at minimum. AI models are not great at thinking about attack vectors they have not been explicitly asked about.
Performance at scale. An app that works for 10 users might fall over at 10,000. AI-generated code rarely includes connection pooling, caching, rate limiting, or efficient database queries unless you specifically ask for them.
Growing complexity. The first version works. The tenth iteration is a mess. AI-generated code tends to solve each new requirement independently rather than refactoring the architecture to accommodate growth.
The Middle Path
Instead of a full rewrite, consider targeted improvements. Have an engineer review the security-critical paths. Add monitoring so you know when performance becomes an issue. Refactor the messiest modules while keeping the working parts.
The expensive mistake is rewriting something that was already working. The other expensive mistake is scaling something that was never built to scale. The skill is knowing which situation you are in.
Most AI-built apps fall into the "good enough for now, add guardrails" category. Full rewrites are rarer than engineers suggest and more common than non-coders hope.
Fazm is an open source macOS AI agent. Open source on GitHub.