核心信息
一位开发者注意到，AI编程代理在执行任务时经常改动测试，而好的测试很少变化——这相当于同一份代码写了两遍。他打算尝试禁止代理创建测试，看看省下的时间能否用于产品开发。
要点
- 好的测试通常很稳定；代理频繁改动测试可能意味着重复劳动。
- 代理可能实际上把同一套逻辑写了两遍：实现一遍、测试一遍。
- 作者正考虑禁止代理生成测试，把节省的时间投入到产品开发中。
AI代理频繁改写测试，可能是个危险信号
原始推文
i haven't sat down and thought about this but something def smells weird here good tests rarely change. but agents seem to constantly be changing tests as they do things they're basically writing the same code twice. this probably is not worth much
引用推文
I’m about to start banning agent’s from creating tests and see how it goes Maybe the time saved from writing tests can be used to actually develop the product