AI代码重写暴露深层思考缺失，DHH对比测试揭示多语言实现差异

DHH利用AI代理将Campfire Once从Ruby on Rails改写为Rust、Elixir和Go多种编程语言，但性能测试结果表明仅依赖AI而忽视代码理解存在严重隐患1。

在可靠性方面存在明显分化1。Rust版本在重载条件下的新消息通知成功传递率仅为1%，远低于Elixir版本的100%1。即便在100 POSTs/s的恒定压力下，Rust客户端接收事件比例约为14%，而Elixir约为60%1。经初步调优，Rust版本broadcast channel容量从256扩大至16384后，传递率提升至90%，但这一改动引发新问题——最大延迟从11秒激增至超过130秒1。相比之下，Elixir版本在相同压力测试中内存使用达1.8GB1。

在吞吐量与错误处理的权衡上也存在差异1。Rust版本处理的请求量约为Elixir版本的4倍，且在该流量级别无HTTP错误出现；但Elixir版本则有约23%的HTTP POST请求超时1。DHH在总结中强调，"你仍然必须批判性思考，知道自己在做什么很重要"1。这一观察表明，盲目使用AI生成代码而不深入理解技术权衡，容易导致性能与可靠性的隐患。

DHH recently used AI agents to rewrite Campfire Once from Ruby on Rails into Rust, Elixir, and Go, revealing significant performance trade-offs that underscore the limits of relying solely on automated code generation.1

The rewrite exposed stark differences in reliability under stress. In heavy load testing, the Rust version achieved only a 1% success rate for new message notifications, whereas the Elixir version maintained 100% delivery.1 Under a constant rate of 100 POSTs per second, the Rust client received approximately 14% of events compared to Elixir's 60%.1 Although the Rust implementation handled roughly four times more requests overall, its message passing performance was severely compromised.1

At the same traffic level, the Rust version recorded no HTTP errors but Elixir timed out approximately 23% of HTTP POST requests.1 The underlying issue traced back to the Rust version's broadcast channel capacity of 256; expanding it to 16,384 improved delivery rates from 14% to 90%, yet maximum latency increased from 11 seconds to over 130 seconds.1 Meanwhile, Elixir's pressure testing at 100 requests per second resulted in memory consumption reaching 1.8GB.1

The exercise illustrates a fundamental principle: "You still have to think critically. It's good to know what you're doing."1 The performance disparities demonstrate that AI-generated code requires rigorous validation and human judgment to navigate the inherent trade-offs between throughput, latency, and reliability across different technology stacks.