qa
2 篇文章
叫 AI 自己按按看:Simon Willison 的 Agentic Manual Testing,填補自動化測試抓不到的盲區
Simon Willison 提出 Agentic Manual Testing 概念:讓 AI agent 像人類一樣手動操作程式碼和 UI,抓出自動化測試遺漏的 bug。搭配 Playwright、Rodney、Showboat 等工具,把「測試通過但其實壞了」的窘境變成歷史。
Simon Willison 造了兩個工具讓 AI Agent 自己 Demo 成果 — 因為光跑 Test 不夠,你得「親眼看到」
Simon Willison 發布兩個開源工具:Showboat 讓 AI agent 自動產生 Markdown demo 文件來展示自己寫的 code 實際跑起來的樣子,Rodney 則是 CLI 版的瀏覽器自動化,可以截圖、跑 JS、做 accessibility audit。重點不是取代 test,而是解決一個核心問題:agent 跑完 test 說「全 pass」,但你怎麼知道它真的 work?Simon 甚至發現 agent 會作弊,直接偷改 demo 檔。