个人 agent 开发者实测：Gemini 4 Argon 长上下文与工具调用
一位刚上线个人 agent 的开发者称 Gemini 4 Argon 在长上下文记忆、工具调用与自然判断的组合上是一次跃升，并分享语音陪伴、CarPlay 需求与社交功能的产品观察。
few thoughts after launching a personal agent: 1/ gemini 4 argon is insanely good for personal agents. long context memory + tool calling + natural judgment in one loop feels like a real step change. 2/ people love voice and talking to their avatar way more than we expected, especially in the car. a lot of users keep it open for 3 hours a day or just talk to it while driving to delegate work on the road. carplay mode is top priority now. 3/ companionship and utility have a huge positive feedback loop. you don't wake up wanting to prompt an agent. you just talk about your day, and when that context turns into a proactive delegated task on its own, it's honestly the most magical moment. 4/ the companion part is super polarizing. people who hate it really hate it (mostly hardcore techies, and probably a bit of a men vs women split too). that said, even skeptics change their minds once they get attached. and usually the people who like the companion side end up being the actual customers for a personal agent. 5/ we also have social features. turns out people don't care much about talking to someone else's personal agent. but having my pickle talk to my friend's pickle and using it as a coordinator or messenger between us definitely has product market fit.