AI Rewrites Code: Why You Still Need to Think

DHH, creator of Ruby on Rails, experimented with AI-generated code rewrites in various languages like Rust, Elixir, and Go. The results highlighted significant issues, such as non-functional differences and performance problems, when relying solely on AI without human oversight. The rewrites demonstrated that AI-generated code often makes arbitrary decisions due to vague prompts, leading to inconsistencies in areas like backwards compatibility and system constraints. For example, the Rust rewrite dropped CSRF tokens and Redis, while the Elixir version maintained closer compatibility with the original Rails code. Performance issues were also evident, with the Rust version suffering from poor concurrency handling and the Elixir version experiencing high latency under load. The article emphasizes that coding is far from being ‘solved’ by AI, and developers still need to understand and critically evaluate the code they work with.