个人助理类产品从平庸到惊艳的跃迁
作者对比 SaaS 的线性进步，认为 Instinct、Muse 这类个人助理在开发中跨过某个阈值后突然「变魔法」，靠的是模型提升加大量试错。
I find the pattern of building in the AI world to be so fascinatingly different from SaaS. In SaaS, product progress felt quite linear. Talk to customers. Get requirements. Build. Validate. Do it again. Every week it got a little bit better. But at no point in Gainsight history did it feel like an "aha moment." But now, when you look at products, it feels like there is an activation threshold where it goes from "meh" to "magic." My understanding is the personal assistant agents (eg instinct, muse) went through this in development. Things didn't really work and all of a sudden they did. Obviously the same thing happened with coding agents late last year. Some of it is model improvements, of course. But a lot is just experimenting to see what will work. Thinking about different use cases that were impossible before. Trying different technical approaches. Basically - insane curiosity. The meh to magic cycle is pretty incredible.