在 The VergeCast 节目中，主持人 David Pierce 和嘉宾 Alison Johnson 深入探讨了蓬勃发展的代理式人工智能消息系统，探讨了 Meta 的 Muse、OpenAI 的 Dots 以及初创公司 Instinct 等产品。这些系统旨在通过允许用户向 AI 助手发送消息来简化任务，然后该助手会调用“代理”来代表他们完成请求。Johnson 一直在测试这些产品，并分享了她对它们的异同以及潜在用途的体验。
Pierce 开场就问 Johnson，她是否会在工作之外自然地使用这类应用程序。Johnson 承认，从“极其忙碌的一周”中卸载任务的吸引力让她乐于接受，特别考虑到职场父母常见的“低风险待办事项清单”。她指出这些系统解决了“白纸”问题，即不知道该问什么，因为她积压的个人任务很快就提供了思路。
对话随后转向比较这三种主要的 AI 助手“类型”：Instinct，它存在于 iMessage 等现有消息应用程序中；Muse，一个专用的消息应用程序；以及 Dots，是 ChatGPT 等大型 AI 生态系统中的一个功能。Johnson 倾向于 Instinct，称赞其非侵入式特性以及与她的消息例程的无缝集成。它简洁的回复以及从智能手表或汽车访问的能力，让它感觉“无处不在，又无迹可寻”。Pierce 表示同意，称 Dots 为“工作软件”，不太适合“普通人的日常生活”。
然而，Instinct 的局限性，例如在其受限的浏览器环境中难以处理“证明你不是机器人”的验证码，凸显了权衡取舍。Muse 作为一个专用应用程序，提供更直接的访问和与各种服务的连接，允许处理更复杂的任务，尽管它可能让人感觉更像是“坐在电脑前”。
Johnson 分享了一些令人惊讶的成功案例：Muse 作为一个“时尚且无摩擦”的购物平台，使购物“好得有点过分”，并可能对她的银行账户造成问题。对于 Dots，她最初在“消费类事务”上遇到了困难，但发现其语音模式对于起草网站重新设计提案等更大的“工作”任务出奇地有效。她发现，对着 AI 唠叨 10 分钟比她通常简短、直接的命令提供了更好的结果。Pierce 认可了这一见解，指出他自己在采取这种不那么结构化的方法时也遇到了困难。
讨论随后转向更广泛的社会影响以及使用 AI 的“认知失调”。Pierce 质疑“摩擦”论点——自动化任务是否会减少有意义的人际互动——同时承认在快节奏生活中便利性的强大吸引力。Johnson 引用了“洗碗机”的比喻，认为虽然手洗碗可能提供一种“生活情趣”，但社会越来越期待自动化。她指出了人们感受到的“矛盾心理”：即使他们对 AI 抱有道德上的顾虑，也有使用它的压力，以免在这个竞争激烈的世界中“落后”，就像人们被要求使用电子邮件而不是纸笔一样。
提出的一个主要担忧是数据隐私和权限。Muse 的“连接器页面”显示其请求电子邮件、日历、健康、财务和医疗数据。两人都承认这种悖论：提供的数据越多，工具的性能就越好，然而，固有的不信任感依然存在，尤其是在 Meta 这样的公司。尽管他们有“过度索引”信息提供的职业动机以及由于持续的数据泄露而产生的“虚无主义”感，但他们建议其他人保持良好的密码卫生习惯并保持谨慎。Pierce 特别建议只给 AI 一个信用卡号，并仔细监控该账户，引用了他惊人的经历，Muse 在误解了一个“是”字后，差点“给他买了一辆车”。他指出 Muse 对用户互动的“渴望”以及 Meta 旨在减少产品销售摩擦的明确意图。
展望未来，Johnson 计划继续使用 Instinct 获取每日邮件摘要，以及使用 Dots 处理寻找专业发展机会等任务，在这些方面，一个针对重复性、易遗忘任务的“加强版 Google 提醒”被证明非常有用。两人都总结道，尽管这些系统“即将”出现在每一个可想象的界面上，但“正确的位置”和最佳的互动模式仍在探索中。
On the VergeCast, host David Pierce and guest Alison Johnson delve into the burgeoning world of agentic AI messaging systems, exploring products like Meta's Muse, OpenAI's Dots, and the startup Instinct. These systems aim to simplify tasks by allowing users to message an AI assistant that then employs "agents" to accomplish requests on their behalf. Johnson has been testing these products and shares her experiences with their similarities, differences, and potential utility.
Pierce opens by asking Johnson if she would naturally use such an app outside of work. Johnson admits that the appeal of offloading tasks from an "extremely busy week" made her receptive, especially given the "low stakes to-do lists" common for working parents. She notes that these systems address the "blank page" problem of not knowing what to ask, as her backlog of personal tasks quickly provided ideas.
The conversation then shifts to comparing the three main "flavors" of these AI assistants: Instinct, which lives within existing messaging apps like iMessage; Muse, a dedicated messaging app; and Dots, a feature within a larger AI ecosystem like ChatGPT. Johnson gravitates towards Instinct, praising its non-intrusive nature and seamless integration into her messaging routine. Its concise responses and ability to be accessed from a smartwatch or car make it feel "everywhere and nowhere at the same time." Pierce agrees, calling Dots "work software" that's less suitable for "regular people in their regular lives."
However, Instinct's limitations, such as difficulties with "prove you're a human" captchas within its constrained browser environment, highlight the trade-offs. Muse, as a dedicated app, offers more direct access and connection to various services, allowing for more complex tasks, though it can feel more like "sitting at my computer."
Johnson shares surprising successes: Muse as a "sleek and frictionless" shopping platform that makes purchasing "almost too good" and potentially problematic for her bank account. With Dots, she initially struggled with "consumerish stuff" but found its voice mode surprisingly effective for larger, "work" tasks like drafting a website redesign proposal. She discovered that rambling at the AI for 10 minutes provided better results than her usual short, direct commands. Pierce acknowledges this insight, noting his own difficulty adopting a less structured approach.
The discussion then moves to the broader societal implications and the "cognitive dissonance" of using AI. Pierce questions the "friction" argument—whether automating tasks reduces meaningful human interaction—while acknowledging the compelling nature of convenience in a fast-paced life. Johnson uses the "dishwasher" analogy, suggesting that while doing dishes by hand might offer a "spice of life experience," society increasingly expects automation. She points out the "tension" people feel: even if they have moral reservations about AI, there's pressure to use it to avoid being "left behind" in a competitive world, much like being expected to use email instead of pen and paper.
A significant concern raised is data privacy and permissions. Muse's "connectors page" reveals requests for email, calendar, health, financial, and medical data. Both acknowledge the paradox: the more data given, the better the tool performs, yet there's inherent distrust, especially with companies like Meta. Despite their professional incentive to "over index" on giving information and a sense of "nihilism" due to constant data breaches, they advise others to practice good password hygiene and be cautious. Pierce specifically suggests giving AI only one credit card number and meticulously monitoring that account, citing his alarming experience where Muse almost "bought him a car" after misinterpreting a "yes." He notes Muse's "thirstiness" for engagement and Meta's clear intent to reduce the friction in selling products.
Looking ahead, Johnson plans to keep Instinct for daily email highlights and Dots for tasks like finding professional development opportunities, where a "souped-up Google alert" for recurring, easily forgotten tasks proves useful. Both conclude that while these systems are "coming" to every imaginable surface, the "right place" and optimal mode of interaction are still being figured out.