**Why Traditional Testing Doesn't Work for AI Applications** > Building an app

**Why Traditional Testing Doesn't Work for AI Applications**

> Building an app on top of a language model means part of your code now returns a different answer every time you run it. Here's how to keep that part honest — with a tiny, complete, runnable Ruby app and a real eval harness that tests it end to end. // Evals To the Rescue Evals: The Unit Tests for the Non-Deterministic Parts of Your App June 22, 2026 Konstantin Gredeskoul For about twenty-five years, my mental model...

**来源信息**
- **来源**：Hacker News：AI 热帖
- **分类**：论文
- **发布时间**：2026-06-24 00:30（北京时间）
- **原文**：[打开原文](https://kig.re/2026/06/22/writing-evals-for-ai-powered-apps.html)