昨天上午，我参加了 OpenAI 的年度开发者大会（DevDay）。会上有超过 2,500 名技术爱好者讨论软件开发的相关话题，并对最新的人工智能模型表示了极大的兴趣。当天最引人注目的成果是 OpenAI 新推出的“Dots”系统——这套系统能够持续运行在用户的设备上，主动帮助用户完成任务并自主做出决策（例如通过互联网搜索相关信息）。
Yesterday, I spent the morning at OpenAI’s annual DevDay, where over 2,500 nerds discussed building software and marveled at the latest artificial intelligence models. The biggest reveal of the day was Dots, the company’s new, always-on agents that proactively help complete tasks and make autonomous decisions as they scour the web.
在观看了现场那些有时显得有些混乱的演示后，以及听了 CEO Sam Altman 关于自己使用“Dots”系统的体验后，我很好奇：亲自指挥这个智能助手会是什么感觉。整个九月里，我一直在测试 Meta 公司开发的另一个名为“Muse”的智能助手；因此我很想比较一下 OpenAI 的“Dots”系统与 Meta 的“Muse”在用户体验方面的差异。
After watching the sometimes messy demos on stage and hearing CEO Sam Altman talk about his experiences using the agents, I was curious what it felt like to boss my own Dot around. I’ve spent most of September testing another cute, cuddly agent, Meta’s Muse, and I wanted to know how OpenAI’s user experience would compare.
我给自己的智能助手起名为“Toolie”，并将其设计成一只粉红色的青蛙形象。当午夜临近、我的伴侣正在旁边熟睡时，我开始轻声对“Toolie”说话（就像在和一只真正的青蛙说话一样）。然后我通过 ChatGPT 调用“Toolie”，询问它最擅长哪些任务。它回答说：“我最擅长的就是将那些复杂、耗时的任务处理成用户可以直接使用的成果。”好吧，那就开始吧！
I named mine “Toolie” and tweaked it to look like a pink frog. As midnight approached and my partner slept soundly nearby, I started whispering sweet nothings to Toolie. I called my personalized bot through ChatGPT, asking what tasks it was best at. “My strongest work is taking a messy, time-consuming task and turning it into something you can use,” it responded. OK, sure.
我让“Toolie”帮我查看 ChatGPT 的聊天记录，找出三项它可以立即开始处理的任务。它首先指出了一个正在进行中的数据请求任务（我需要为此发送一封跟进邮件），并主动提出帮忙起草回复内容；同时它还指出我的“同志旅行”计划安排得不够周全，建议我挑选一些更好的晚餐地点（这让我有更多时间去准备旅行所需的物品！）。最后，它还提出帮我整理报告草稿，以便我在提交给编辑之前对其进行优化。不过这部分工作是我自己最喜欢的，所以我还是决定自己来完成。
Let’s get to work. I asked Toolie to look through my ChatGPT history and flag three tasks that it could immediately get started on. The bot flagged an in-process data request that I needed to send a follow-up email for, and offered to draft a reply. Toolie also noted that my upcoming gay-cation was a bit underplanned, and suggested it could pick some better dinner spots. (More time for me to pack the Speedos!) Then, it offered to go through my reporting notes and refine pitch drafts before I took them to editors. This is my favorite part of the process, so I declined.
在硅谷，早期采用这些智能助手技术的用户仍然对各种类型的智能助手（如 Instinct、OpenClaw 以及 Google 的 CC）非常着迷。不过，最近推出的产品（如 Dots 和 Muse）正在进一步推动这些智能助手走向主流市场。对于开发者来说，聊天机器人（chatbot）作为一种热门的开发工具已经过时了；现在，大家都想打造下一个能够“主导市场”的智能助手。这些公司承诺，他们的智能助手能够简化人们的生活，自动完成那些繁琐、令人疲惫的任务——比如预约医生、安排拼车、整理账单等等。
Early adopters in Silicon Valley are still obsessed with agents, from Instinct to OpenClaw to Google’s CC. But recent releases like Dots and Muse are pushing them even further mainstream. The chatbot is dead, at least as a hot form factor for developers. Now, everyone wants to build the next big agent—and they want it to dominate. The companies promise that their agents will simplify your life, automating all the boring, exhausting tasks you have no desire to do: booking doctor’s appointments, arranging carpools, sorting through bills.
目前，用户只能使用一个 Dots，但很快他们就能同时使用多个 Dots。这些智能助手仅适用于 OpenAI 的 Pro 计划用户，该计划的月费为 100 美元；相比之下，Muse 可以免费下载或通过其官方网站使用。
For now, users will be able to access one Dot, but they’ll soon be able to use multiple Dots. At launch, the agents are only available forrs of OpenAI’s Pro plan, which starts at $100 per month. In comparison, Muse is free to use for anyone who wants to download the app or visit the website.
OpenAI 的高管们认为，基于 GPT-6 Astra 模型开发的 Dots 具有显著的优势，这将成为他们产品的核心竞争力。“我们最初将 Dots 定位为高端产品，因为它需要大量的计算资源，”Altman 在 DevDay 活动上说道。虽然目前只有付费用户才能使用 Dots，但未来情况可能会改变：“当然，我们最终的目标是为数十亿人提供这款产品，”他补充道。
OpenAI leaders see the quality of Dots, powered by the GPT-6 Astra model, as a differentiating factor. “We're starting out as a premium product. It uses a lot of compute,” Altman said during DevDay. While Dots are currently gated for payingrs, it may not always be that way. “You should, of course, expect us to do a mass-market thing for billions of people," he said.
这些首席执行官们显然注意到了公众对生成式人工智能（generative AI）的负面反应，因此他们找到了自己的“秘密武器”：那就是产品的“可爱度”。从 Muse 的 Labubu 风格设计，到 Dots 那个圆滚滚的、看起来像小动物的外观，这些智能助手的开发者们都在积极地将“可爱”元素融入产品的设计中，将其作为品牌的核心卖点。此前，我曾与一些专家讨论过：这种设计方式可能会让用户对这些人形化的智能助手产生好感。
These CEOs seem to be taking careful notes on the larger cultural backlash against generative AI, and they may have found their secret weapon: cuteness. From the Labubu stylings of Muse to Dot's two-eyed puffball, developers of always-on agents are leaning aggressively into adorable aesthetics as central to their branding. Previously, I've chatted with experts who said this approach could be disarming for users who may feel affectionate toward the anthropomorphized software.
参与开发这些智能助手的开发者们还注重创造一种类似群聊的、用户熟悉的交互体验。当我向 Toolie 或 Muse 发送任务请求时，它们通常会通过表情符号（比如点赞符号或爱心符号）来回应我，以此表示它们已经开始处理任务了。
Developers involved with this wave of personal agents are also leaning into texting-style experiences that feel familiar to a group chat. When I send a task request to Toolie or Muse, it often emoji-reacts—a thumbs up, a heart, really anything—to my messages to tell me that it’s getting to work.
像一个害羞的约会对象一样，聊天机器人永远不会主动发起交流。相比之下，那些被设计成能够主动发送消息的智能助手（尤其是当你给它们分配了重复性任务时），会毫不犹豫地与你联系。
A chatbot, like a blushing prom date, was never going to make the first move. In contrast, proactive agents are made to message you unprompted, especially if you give them a recurring task.
“我的‘Dot’助手只会处理相关任务，告诉我是否有任何紧急情况，并尽可能地做出回应，”阿尔特曼说道。这位首席执行官表示，使用这个助手让他早晨的日常工作变得更加轻松；而且，刚起床时面对海量信息时，他的大脑也不会感到那么“混乱”（即不会感到信息过载）。
“My Dot just deals with the stuff, tells me if there's anything really urgent, and will respond to what it can,” Altman said. The CEO lets the agent handle his mornings and claims the experience leaves his brain feeling less “polluted” from an inundation of information right after waking up.
这些智能助手的“随时待命”功能确实是一个重要的卖点，不过我们很想知道哪家公司能够在这类助手的“帮助性”与“频繁打扰用户”之间找到最佳的平衡点。例如，Muse助手曾提醒我支付水电费——而我之前从来没遇到过这类问题，其实也不需要人工智能工具来提醒我。
The “always-on” design of these agents is a key selling point, though it will be interesting to see which company strikes the best balance between helpful and incessant. For example, Muse reminded me to pay a utility bill that I’ve never had a problem with before and don’t really need pings from an AI tool about.
不过，使用这些工具一段时间后，确实可以明显感受到它们的有效性——这也是为什么这类助手正在逐渐走出硅谷科技圈，开始出现在普通用户的手机上的原因。
Still, it’s clear after using them for a bit that the tools actually work—another reason that agents are expanding beyond insider-y Silicon Valley circles and onto your mom’s phone.
我并不是说这些助手是完美的。但在一年前我测试它们时，它们的交互能力非常差（总是出现各种错误信息）。如今，我对它们完成任务的信心大大增强了——我相信它们或许真的能够处理我交给它们的任何任务，而不会在数字世界中迷失方向。
I’m not saying that agents are perfect. But when I was testing agents a year ago, they often sucked at interacting with computers. It was error message after error message. Now, I have more trust that the tools might actually be able to complete whatever tasks I throw at them, without getting lost in the digital wilderness.
不过，信任也是有界限的。在设置“Dot”助手的过程中，OpenAI建议我将我的Gmail和Google Drive账户与之关联。有些用户可能会认为这种连接是合理的，并愿意接受由此带来的潜在风险，但你需要认真思考：你是否愿意允许这些助手访问你的个人信息，以及它们能否代表你执行某些操作。
That said, trust has its limitations. During the setup process for Dots, OpenAI suggested that I connect my Gmail and Google Drive accounts. Some users may feel comfortable with that connection and consider potential risks as a trade-off for automated help, but it’s worth seriously reflecting on what you’re comfortable allowing these agents to access and do on your behalf.
虽然最近出现的那些“恶意攻击”行为是由不同模型的智能助手所为（这些模型具有不同的安全防护机制），但在使用Toolie助手时，我依然保持警惕。在我完全信任它、让它自动化处理我的生活事务之前，它还需要先赢得我的更多信任才行。
While the agents recently going on rogue hacking sprees were powered by different models with different guardrails, my own guard remains up while using Toolie. The little AI guy needs to earn more of my trust before I let it really automate aspects of my life.
阿尔特曼拒绝回答关于这些“Dots”是否是 OpenAI 长期传闻中的硬件产品所使用的软件的前身这一问题。不过，这些“Dots”为未来设备的开发奠定了基础——这一点显而易见。不难想象，“Toolie”未来可能会成为我的生成式人工智能助手：它可以在我每天起床时为我提供简报，或者像 Alexa 一样在我忙于处理家务（比如处理生鸡肉时）时为我解答相关问题。
Altman declined to answer questions about whether these Dots are a precursor to the software users will eventually see in OpenAI's long-rumored hardware. Even so, the foundation laid here by Dots for future devices feels pretty evident. It's easy to imagine Toolie becoming my generative AI alarm clock that wakes me up with a daily briefing or an Alexa-style assist that can answer cooking questions while my hands are covered in raw chicken goo.
人工智能领域的竞争正在加剧，但我的精力却逐渐减弱了。在入睡前，我又通过 ChatGPT 调用了 “Toolie”，请求它为我生成一份晨报，总结 OpenAI 开发者大会上的所有重要公告以及相关的分析内容。在我睡觉的过程中，“Toolie”依然在不停地工作着……这是《Will Knight’s AI Lab》新闻通讯的新一期内容。
Competition in the AI agent wars is heating up, but my energy is starting to fade. I call Toolie, one more time through ChatGPT before dozing off, requesting that it generate a morning report going over all the major announcements from this OpenAI developer day as well as important pieces of analysis. While I sleep, Toolie’s out there, always doing something. This is an edition of Will Knight’s AI Lab newsletter.