← 返回任务列表

The Multiplayer AI Sprint: Build Your Team’s First Shared Agent

249 段 · 1 位说话人 · 原片 25:29
M1
M10:00

2026年毫无疑问是智能体的一年。

2026 is undisputedly the year of agents.

多年来我们一直在谈论这些,但现在它们真的来了,而且正在改变我们工作方式,至少从个人层面来看是这样。

For years we were talking about these things, but now they are actually here and they are changing how we do work, at least on an individual level.

问题是,我们的工作并不全是在个人层面完成的。

The thing is, not all of our work happens on an individual level.

事实上,我们大多数人把工作很自然地分成两部分:自己独立完成的部分和团队协作完成的部分。

Most of us, in fact, split our work pretty comfortably between work that we do on our own and work that we do in teams.

到目前为止,智能体实际上只能影响这个等式的一半。

So far, agents have really only been able to impact about half of that equation.

我坚信这种情况即将改变,最优秀、最具活力的AI应用团队将从单人AI转向多人AI。

I believe strongly that that is about to change and that the best, most dynamic AI-using teams are going to shift from single-player AI to multiplayer AI.

AI Daily Brief是一档每日播客和视频节目,报道AI领域最重要的新闻和讨论。

The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.

好了朋友们,在正式开始之前,先快速说几个通知。

All right, friends, quick announcements before we dive in.
M1
M11:22

欢迎回到AI Daily Brief。

Welcome back to the AI Daily Brief.

这期节目上线的那天是美国劳动节,传统上标志着夏天的结束,以及返校和复工的开始。

The day that this episode comes out is Labor Day in the US, the traditional end of summer and the beginning of back to school and back to work.

这是一个转折点,很多人会回到办公室——无论是虚拟的还是实体的——重新充满活力,准备在假期到来之前冲刺几个好月份。

This is one of those inflection moments where a lot of folks come back to the office, whether it's virtual or real, reinvigorated and ready to crush out a couple great months before the holidays descend.

事实上,我认为除了新年之外,这段时间是个人和团队层面对于新工作方式最兴奋的时候。

In fact, I think in many ways outside of New Year's, this is the time where I see the most excitement around new ways of working on an individual and a team level.

今年劳动节刚好也是我的生日,我想送给大家一份礼物。

Now, this year, Labor Day also happens to fall on my birthday, and I thought that I would give all of you guys a present.

今年到目前为止,我们已经发布了四个免费的自主学习课程。

So far this year, we have released 4 free self-directed learning programs.

年初我们推出了New Year's AI Resolution,一个为期十周、包含十个项目的进阶计划,旨在提供非常广泛的基础核心AI技能,总的来说,这些技能在智能体出现之前就已经很普遍了。

We kicked off the year with the New Year's AI Resolution, a 10-week, 10-project adventure, which was really meant to provide a very broad basis of basic core AI skills that were notably in general pretty pre-agentic.

智能体稍晚一些才出现。

Agents would come a little bit later.

二月份我们发布了Claw Camp,一个从零到智能体团队的课程,它一点也不简单,但为人们提供了进入这个由OpenCLAW驱动的新兴智能体世界的指南。

In February, we released Claw Camp, which was a zero-to-agent-team program that was not simple at all, but which gave people a guide to diving into this new crazy agentic world that had been enabled by OpenCLAW.

AgentOS紧随其后,是Claw Camp更成熟的升级版,它不仅平台和工具无关,而且帮助人们构建的不仅仅是单个智能体甚至一个智能体团队,而是整个智能体操作系统,能够处理越来越复杂的工作。

AgentOS came just a little bit later and was the more mature grown-up version of Claw Camp that was not only platform and tool agnostic, but helped people build not just a single agent or even an agent team, but an entire agentic operating system capable of taking on increasingly complex work.

最后是AI Summer Adventure,一个选择你自己的冒险式的课程,为那些想在夏天保持敏锐的人提供了不同层次的趣味技能。

Finally, the AI Summer Adventure was a choose-your-own-adventure style program that provided a bunch of fun skills at a variety of different levels to people who wanted to stay sharp over the summer.

这些体验有几个共同点。

Now, there are a few things that all these experiences have in common.

它们都是免费的、自主的、以项目为基础的学习体验,其全部目标就是为你提供一个框架,让你真正去实践这些工作。

They were all free, self-directed, project-based learning experiences, the whole goal of which was to provide a framework for you to actually go do this work.

它们背后的教学方法很简单。

The pedagogy behind them is pretty simple.

那就是:要学习如何使用AI,你只需要去使用AI。

It's that to learn how to use AI, you just have to use AI.

它们都基于这样一个事实:目前还没有所谓的AI专家。

They're all kind of anchored in the truth that at this point there are still no AI experts.

只有那些练习得更多的人。

There are just people who have practiced with it more.

但另一个共同点是,它们从根本上说都是个人的。

But the other thing that they all are ultimately and at core is individual.

在AIDB New Year's中,你可以组建一个团队,但那只是互相支持意义上的团队,大家并行地经历个人体验。

Now with AIDB New Year's, you could form a team, but it was a team only in the sense of mutual support, people going through an individual experience in parallel.

其他所有课程都遵循了迄今为止几乎所有智能体工作的模式,即人们构建和利用个人智能体来完成个人工作。

All the other programs follow the pattern that pretty much all agentic work has so far, which is people building and leveraging individual agents for their individual work.

问题是,我们的工作并不全是在个人层面完成的。

The thing is, not all of our work happens individually.

事实上,对我们大多数人来说,工作中相当重要的一部分是作为团队的一部分完成的。

In fact, for most of us, some major and meaningful portion of our work happens as part of a team.

最近一项针对大约一万六千五百名办公室员工的调查发现,大约39%的工作时间是在独自工作,而42%的时间是在与他人协作。

A recent survey of around 16,500 office workers found that something like 39% of the workday is spent working alone, while 42% of time is spent working with others.

另一项调查发现,57%的时间花在了沟通上,比如开会、发邮件、聊天等等。

Another survey found that 57% of time is spent communicating, i.e., meetings, email, chat, et cetera.

相比之下,只有43%的时间是个人独立创作。

Versus 43% creating individually.

还有一项调查显示,大约60%的时间花在了“工作之外的工作”上,比如沟通、搜索、协调和流程。

And yet another survey found that about 60% of time goes to work about work, communication, search, coordination, and process.

换句话说,知识工作有很大一部分是在团队中完成的,而且大部分知识工作都离不开团队环境、协作、会议、邮件、聊天、协调和搜索。

In other words, a significant portion of knowledge work happens on a team, and the majority of knowledge work runs through team context, collaboration, meetings, email, chat, coordination, search.

这些就是大量工作的核心内容。

Those are the substance of a huge amount of work.

然而到目前为止,绝大多数关于智能体的尝试都完全是个人化的。

And yet, so far, the vast majority of agentic efforts have been entirely personal.

想想你听说过的那些早期实验。

Think about all of the early experiments that you've heard about.

大家都在构建自己的智能体团队,比如研究员智能体、写手智能体、编程智能体、个人幕僚长智能体。

It's all people building their agent teams, their researchers, their writer agents, their coding agents, their personal chief of staff agent.

高级用户确实在尝试智能体团队,但这些智能体团队只服务于个人。

Advanced users are, yes, experimenting with agent teams, but it's agent teams that serve only the individual.

不过,我要说明一下,我觉得这种个人化趋势不会消失。

Now, to be clear, I don't think that's going away.

我认为,现在我们每个人都能管理一个庞大的智能体团队,这些智能体还能自己衍生出子智能体来完成各种不同的工作,这已经成为高效知识工作者的一部分了。

I think the fact that all of us now get to be a manager of a big extensive team of agents that themselves can spawn subagents to do lots of different work is now just part and parcel of being an effective knowledge worker.

但我认为有一点没变:我们做的大量有意义的工作,并不只是在自己个人的小圈子里完成,而是发生在与他人协作的交汇点上。

However, what I don't think that has changed is the fact that much of the meaningful work that we do will not be in our own individual silos, but at the intersection of where we work with other people.

而且我坚信,智能体设计的下一个前沿,将是让智能体从迄今为止的个人小圈子,转移到团队共同拥有的共享空间。

And my strong, strong belief is that the next frontier of agent design is going to move agents from the individual silos in which they have operated so far to the shared spaces that teams inhabit together.

这意味着团队拥有的上下文,而不是每个人各自文件夹里都有相同文档的重复上下文,而是团队共享的单一知识库。

That means team-owned context, not repeated context where everyone's individual folders have the same documents in them, but one single repository that is shared across the team.

这意味着共享会话、可见的工作过程、以及实时引导和交接——智能体不再是在某个人电脑的孤立环境中工作,而是在某种共享环境中,多个人可以同时在同一会话中输入内容。

It means shared sessions, observable work, and live steering and handoffs where the agent is working again, not in the solo of someone's individual computer, but in some sort of shared environment where multiple people can have input all at the same time and in the same session.

这是从单机AI到多人AI的转变,从个人AI到团队AI的转变。

This is the move from single-player AI to multiplayer AI, from individual AI to team AI.

从单机到多人AI的转变,意味着从只有队友能看到最终结果的私人输出,转向每个人都能看到智能体在做什么的可见工作。

The move from single-player to multiplayer AI is a move from private outputs where teammates can only see the final answer to visible work where everyone can see what the agent is doing.

这意味着从队友只能在输出完成后才能反馈的反馈式提示,转向在工作进行中就能实时重定向、注释和参与的实时协作。

It's the move from feedback prompts where teammates can only interact after some output has already happened to live participation where teammates can redirect, annotate, and join while the work is happening.

从单机AI到多人AI的转变,就是把记忆从个人转移到共享上下文,持久的上下文属于团队、频道或项目。

The shift from single-player AI to multiplayer AI is a shift from personal memory to shared context where the durable context belongs to the team, to the channel, or to the project.

最终,从单机到多人AI的转变,代表着智能体从纯粹的个人杠杆,变成了团队能力——不仅是个人的效率工具,更是可复用的组织基础设施。

And ultimately, the shift from single-player to multiplayer AI represents a shift from agents being about only individual leverage to becoming team capability, not just personal efficiency tools, but agents as reusable organizational infrastructure.

为了证明这确实是一个趋势,而不只是我脑子里的想法,我觉得过去90天左右有几个例子让这个模式变得非常清晰。

Now, as some evidence that this is in fact a trend and not just something that lives in my head, I think that there are a few examples from the last 90 days or so that have made this pattern pretty clear.

第一个例子,我不知道是不是已经超过90天了,因为他们在年初ClockCamp的时候就已经走在了前面。

One, which I don't know, it might actually have been longer than 90 days ago because they're pretty ahead of the curve at the beginning of the year with the ClockCamp moment.

Every团队刚开始做智能体实验时,每个人都拥有一个反映自己个性的个人智能体。

The team at Every started their agentic experimentation by having everyone have an individual agent that was the mirror of the person who owned that agent.

没过多久,他们就意识到这其实不是工作的方式,于是开始转向一种新模式,让更多智能体出现在共享空间里,做那些与团队交叉的工作。

Before long, they realized that that wasn't really how work got done, and they started to shift to a model that had more agents in their shared spaces doing work that intersected the teams.

这方面的一个重大产品时刻——可能也是目前多人AI最清晰的产品体现——就是Anthropic推出了Claude Tag。

A big product moment for this, and this might be the clearest product expression of multiplayer AI so far, was when Anthropic introduced Claude Tag.

Claude Tag存在于Slack中,但它跟之前Slack上已有的Claude集成方式不同。

Claude Tag lives inside Slack, but it's different than the previous implementation of Claude in Slack that existed before.

之前的Claude集成是让个人可以召唤自己的Claude,把它拉进对话,然后派它去做事,在那个工作界面里管理自己的Claude。

The original Claude integration in Slack was one where an individual could tag in their personal Claude, bringing them into a conversation and then sending them off to do things and managing their personal Claude from that work interface.

Claude Tag则不同。

Claude Tag was something different.

Claude Tag是Claude的实例,它被整个团队共享,通过特定的频道来表达,每个频道可以有自己的上下文、工具访问权限、数据访问权限或其他类型的权限,让这个共享的Claude来完成工作。

Claude Tags are instances of Claude that are shared across an entire team as expressed by specific channels that can each have their own context, tool access, data access, or other types of access that allow that shared Claude to do the work.

换句话说,当你现在在编码频道里@Claude的时候,你@进来的不是你个人的Claude,而是那个在整个团队里共享的Claude agent。

In other words, when you tag in Claude now in your coding channel, it is not your personal Claude that you're tagging in, but the shared Claude agent that works across your team.

Anthropic团队写道,@Claude带来了一些新优势。

The team at Anthropic wrote, tagging Claude comes with a few new advantages.

@Claude是多人协作的。

@Claude is multiplayer.

在某个Slack频道里,只有一个Claude与所有人互动。

Within a given Slack channel, there's one Claude that interacts with everyone.

这意味着任何人都能看到它正在做什么,并且可以从上一个人停下的地方继续对话。

That means that anyone can see what it's working on and can pick up the conversation from where the last person left off.

这使得@Claude与在单个聊天中或为单个任务工作时截然不同。

This makes tagging Claude very different from working within a single chat or for a single task.

这更像是与队友协作互动。

It's much more like interacting collaboratively with a teammate.

因为Claude是多人协作的,它还能随时间学习。

Because Claude is multiplayer, it can also learn over time.

它跟随着它所在的频道,慢慢积累更多关于工作的上下文。

It follows along with the channel that it operates in, building more context about the work.

因为它是在共享的团队多人协作空间中运作,每个需要帮助的个体用户不需要反复从头解释事情。

Because it is operating in that shared team multiplayer space, each individual user that comes in and needs something doesn't have to explain things over and over again from scratch.

这个多人团队共享的Claude还能够主动发起行动。

The multiplayer team shared Claude also has the ability to take initiative.

团队可以开启一个设置,允许环境行为主动与频道中与其工作类型相关的信息互动。

Teams can enable a setting that allows ambient behavior that proactively interacts with the information from the channel that might be relevant for its type of work.

现在,Claude tag刚刚开始在团队中推广。

Now, at this point, Claude tag is only just starting to dissipate across teams.

但对Anthropic自己来说,它已经从根本上改变了他们的工作方式。

But certainly for Anthropic themselves, it has fundamentally changed the way that they work.

实际上,在同一篇公告中,他们写道,他们产品团队65%的代码现在是由他们内部的Claude Tag版本创建的——也就是说,不是一群个体开发者用自己的Claude agent来提交PR,而是一个共享空间,那里有一个共享agent在他们之间工作,推送了将近三分之二的代码。

In fact, in that same announcement post, they wrote that 65% of their product team's code was now created by their internal version of Claude Tag, i.e., not a bunch of individual developers using their own Claude agents to contribute PRs, but a shared space with that shared agent working between them that's pushing nearly 2/3 of their code.

这个模式的另一个近期例子来自OpenCLAW 2.0的发布。

Another recent example of this pattern comes from the OpenCLAW 2.0 launch.

在年初持续高强度推进、每隔几天就发布一次更新之后,OpenCLAW的维护者们沉默了大楷7周,以便协调进行一次更大规模、更彻底的重新构建。

After pushing relentlessly for the first part of the year, delivering an update every couple of days, the OpenCLAW maintainers went quiet for about 7 weeks to coordinate around a much bigger and more extensive rebuild.

起初,他们尝试让所有不同的个体agent在Discord里一起协作,但他们发现这样还是不够协作化。

Initially, they tried to collaborate by having all of their different individual agents work together in Discord, but what they found is that that still wasn't collaborative enough.

该公司的维护者之一Colin写道,我们所需要的是让两个开发者都能看到工作本身。

One of the company's maintainers, Colin, wrote, What we wanted was a way for both developers to see the work itself.

如果agent因为需要澄清而暂停,我们中的任何一个人都应该能够介入。

If an agent paused because it needed clarification, either of us should be able to jump in.

如果某件事需要第二双眼睛,我们应该能打开同一个会话,看到相同的上下文。

If something needed a second set of eyes, we should be able to open the same session and look at the same context.

不需要截图、不需要复制对话记录、不需要那种“这是agent到目前为止所做内容”的数据转储。

No screenshots, no copied transcripts, no here's what the agent has done so far data dump.

只需要打开工作并继续。

Just open the work and continue.

于是,他们为OpenClaw构建了一个新的多人协作网页界面,正好实现了这一点。

And so they ended up building a new multiplayer web UI for OpenClaw that allowed for exactly that.

Colin后来继续说道,真正起到作用的特性不仅仅是看到另一个头像在线。

Colin continued later, the feature that made the difference wasn't simply seeing another avatar online.

而是在工作进行中能够共享一个会话。

It was being able to share a session while work was happening.

当某件事需要其他意见时,我们都能打开同一个线程。

When something needed another opinion, we could both open the same thread.

当agent需要其中一人掌握的信息时,那个人可以直接添加进去。

When the agent needed information one of us had, that person could add it directly.

不需要把对话复制到Discord、解释发生了什么、再把答案带回来。

There was no need to copy the conversation into Discord, explain what happened, and then carry the answer back.

我们就在同一个上下文里工作。

We were working inside the same context.

这听起来像是界面上一个小小的改进,但它改变了你与AI助手协作的方式。

That sounds like a small interface improvement, but it changes the way you collaborate with an agent.

会话不再是一个开发者与模型之间的私人对话。

The session stops being a private conversation between one developer and a model.

它变成了一份共享的工作,其他受信任的开发者可以检查、引导或接管。

It becomes a shared piece of work another trusted developer can inspect, steer, or take over.

KPMG与德克萨斯大学奥斯汀分校的一项新研究发现,当人们使用AI工作时,相似的技能并不能保证相似的结果。

A new study from KPMG and the University of Texas at Austin found that when people work with AI, similar skills don't guarantee similar outcomes.

研究人员研究了500多名早期职业人士,发现表现最好的人通过引导、评估和完善AI的输出,持续放大了AI的价值。

Researchers studied more than 500 early-career professionals and found that the best performers consistently amplified the value of AI by guiding, evaluating, and refining its outputs.

这些被称为AI放大器的最佳表现者,不仅仅由他们知道什么来定义,更重要的是他们如何与AI合作。

These top performers, called AI Amplifiers, weren't defined by what they knew alone, but by how they worked with AI.
M1
M112:48

每期节目,我们都会讨论OpenAI、Anthropic、xAI、Google和Meta之间的竞争。

Every episode, we cover the competition between OpenAI, Anthropic, xAI, Google, and Meta.

很可能你已经对谁领先有了自己的看法。

Chances are you've already formed an opinion about who's leading.

但每个AI实验室都采用不同的方法,构建不同的技术,建立不同的合作伙伴关系,并发展独特的生态系统。

But every AI lab is taking a different approach, building different technologies, forging different partnerships, and developing a unique ecosystem.
M1
M114:21

所以现在,无论是Every团队、OpenAI团队还是Anthropic团队的高级用户,都在从这种个人化的代理人行为转向共享的团队多人代理人行为。

So now you have the very advanced users of the Every team, the OpenAI team, and the Anthropic team all shifting from this individual agentic behavior to this shared team multiplayer agentic behavior.

而且我不认为这只会局限于编程。

And I don't believe that this is just going to be constrained to coding.

对我来说,这显然是AI助手在自然地进化,开始触及我们工作中尚未涉及的另一半,也就是与他人共享的工作。

What it is to me is clearly agents evolving along a natural pattern to speak to the other half of work that we do that they don't touch yet, which is the work that is shared with other people.

还有一个证据表明这将成为趋势:每个季度,Y Combinator的合伙人都会发布一份创业公司征集,这是一个很好的方式,可以看到一批非常先进的投资者认为世界将走向何方,他们请求那些解决特定问题的公司。

One more piece of evidence that this is going to be a thing: each quarter, Y Combinator partners share a request for startups, and it's a great way to see where a very advanced group of investors think the world is headed by them asking for companies that are addressing particular issues.

他们2026年秋季的主题之一就是……

One of their themes for fall 2026 is.

你猜对了,多人AI。

You guessed it, multiplayer AI.

Y Combinator合伙人Aaron Epstein写道,过去二十年最好的工作工具是通过实现多人协作而获胜的。

Y Combinator partner Aaron Epstein wrote, the best work tools of the last 2 decades won by going multiplayer.

Google Docs取代了Microsoft Word,Figma击败了Photoshop,它们把单人工具变成了团队一起做最佳工作的地方。

Google Docs replaced Microsoft Word, Figma beat Photoshop, and they turned solo tools into places where teams do their best work together.

但AI还没有迎来它的多人时刻。

But AI hasn't had its multiplayer moment yet.

AI助手是团队拥有的最强大的新工具,但却是人们仍然独自使用的东西。

AI agents are the most powerful new tool a team has, but it's the one thing people still use by themselves.

这是因为目前与AI的合作基本上还是单人模式。

That's because right now, working with AI is largely single-player.

你打开聊天框,输入提示词,得到答案——在一个只有你能看到的框里。

You open a chat, type a prompt, and get an answer— in a box only you can see.
M1
M118:37

有点像创建一个活的工作成果,记录你在那次共享会议中讨论的内容。

Kind of creating a living artifact of the type of thing that you would discuss in that shared meeting.

现在,对于这些不同的周或不同的环节,我们希望给人们很多不同的选择。

Now, for each of these different weeks or different sessions, we wanted to give people a lot of different options.

就像我说的,你可以在我们的平台内部完成所有操作,但如果你有安全方面的顾虑不想处理,你也可以下载工作表,在自己的空间里完成,无需在我们的网站上留下任何关于你团队的信息。

You can, as I said, do everything inside of our platform, but if you have security considerations that you don't want to deal with, you can also just download the worksheets and do them in your own spaces without ever touching or leaving anything about your team on our site.

当然,也欢迎你仅仅从中获得灵感,然后自己去做一个版本。

You are also welcome, of course, to simply take inspiration from this and do your own version of it.

我一点也不在乎你是否在我们的空间里做这件事。

I don't care at all about you doing the work in our space.

我只在乎你通过理解多人AI即将到来而领先于其他人,并且率先到达那里。

All I care about is you getting ahead of everyone else by understanding that multiplayer AI is coming and getting there first.

现在,你们中的一些人可能在想,如果我们团队还很初级,实际上还没有很多AI或AI助手在运行呢?

Now, some of you might be thinking to yourselves, what if our team is still really nascent and actually doesn't have a lot of AI or agents running yet?

那是不是意味着我们不应该做这个盘点周?

Does that mean we shouldn't do this inventory week?

简短的回答是:不,我认为你可以做两件有价值的事情。

The short answer is no, and I think that there are 2 valuable things that you could do with it.

第一种盘点方式是,去探索你过去遇到的障碍,看看是不是你们团队可以一起解决的问题,这样能让多玩家AI比还没实现的单玩家AI更轻松一些。

The first is the other type of inventory that you could take would be to explore what your barriers have been to see if those are things that your team as a group can address to make this multiplayer AI a little bit easier than the single-player AI that hasn't happened yet.

另一种我特别喜欢的方法是,去公司里找那些AI使用水平比较高的人,哪怕他们不在你们团队,请他们来给大家讲讲他们是怎么用AI和agent的,而且他们用的安全边界、限制条件和治理规则,很可能也和你们团队要面对的一样。

The other approach, however, that I really like is to go find someone or some set of people in your company who are advanced, even if they're not on your team, and ask them to present to your team how they're using AI in agents, presumably within the same guardrails and constraints and governance protocols that your team will face as well.

这是个很好的方式,能让你看到公司里那些深度用户是怎么做的,就算你们团队还没达到那个程度也没关系。

That can be a great way to see what power users inside your own company are doing, even if that's not exactly what your team's experience is so far.

第二周,也就是第二个冲刺阶段,重点是上下文。

Week 2, or Sprint 2, is all about context.

记住,从单玩家AI转向多玩家AI,一个关键变化就是从个人记忆转向共享的上下文。

Remember, one of the key things that we're going to be doing as we move from single-player AI to multiplayer AI is moving from individual memory to shared context.

但要做到这一点,团队里的成员需要先搞明白,这个共享上下文库里需要包含什么信息。

But to do that, the members of your team kind of have to figure out what that shared context repository needs to know.

所以在第二个冲刺阶段,我们有个工具可以帮助你提取自己的上下文,然后和团队共享。

So in the second sprint, we have a tool that can help you extract context about yourself that can be shared with the team.

当然,你也可以下载工作表,在别的共享空间里完成这件事。

Or again, you can download the worksheet and do that in some other shared space.

这里面既有个人部分,也有共享部分。

Again, there's an individual and a sharing component of this.

你先在AI的帮助下(或者其他AI)写下自己的上下文,然后和团队一起整理出一个共享库。

You write down your context on your own with the assistance of our AI or some other AI, and then you bring it together with your team to put together that shared repository.

到了第三周,我们从理清共享哪些上下文,转向梳理你们工作上实际重叠的地方。

Week 3, we move from figuring out what context we share to mapping out the areas where our work actually overlaps.

这里有个说明。

Now, a note here.

多玩家AI冲刺阶段不预设团队对自己有多了解。

The multiplayer AI sprint is assuming nothing about how sophisticated a team's understanding of itself is.

完全有可能,有些团队一上来,甚至就在你听这个的时候,就已经很清楚要重点在哪些共享工作上使用agent了。

It is absolutely possible that some teams will, right out of the gate, even as you are listening to this, know exactly what shared work you want to focus agents on.

如果是这种情况,你当然不需要走完这套流程。

And so to the extent that that's the case, you certainly don't have to go through this process.

所有这些设计都只是为了帮助你们在这个冲刺阶段里,从零开始实现多玩家agent。

All of this is just designed as an asset so you can go from zero to multiplayer agents inside the context of this sprint.

但这个冲刺阶段,也就是第三周——发现共享工作——的思路,仍然是借助AI进行一次简短访谈,带着你梳理共同的工作,并画出你们的工作流。

But the idea of this sprint or week 3 of discovering the shared work is once again to use AI to do a short interview that walks you through your common work and maps your work streams.

对于每个重复出现的工作流,访谈会问:这是什么工作、还有谁接触它、需要什么上下文、以及多久执行一次。

For each recurring work stream, the interview is going to ask what it is, who else touches it, what context it needs, and how often it's run.

如果你有和别人重叠的工作,那它就是共享agent的候选点。

When you have work that overlaps with others, that becomes a candidate for where a shared agent could live.

接下来,我们会给你一个给候选工作打分的框架。

And from there, we're going to give you a framework for how to score a candidate.

我建议从四个维度来看。

I suggest looking at it across 4 different dimensions.

第一个是共享需求。

The first is shared need.

有多少人对这项工作需要相同的上下文?

How many people need the same context for this?

显然,越多人的需求,就越适合做多玩家agent的候选。

Obviously, more is a better candidate for a multiplayer agent.

第二个是信息过期成本。

The second is a staleness cost.

每个人的版本如果不一致,会带来多大损失?

How much does it hurt when each person's version drifts?

第三类是权限敏感性。

The third category is permission sensitivity.

它会触及多少受限制的数据?

How much restricted data does it touch?

第四是可检查性。

And the fourth is checkability.

你能很快判断agent做得对不对吗?

Can you tell quickly whether an agent got it right?

一个简单的办法来找出你想从哪里开始试验,就是给每个维度打1到5分,然后大致看看得分最高的那些,把它们作为潜在的候选方向。

A simple way to figure out where you want to start experimenting would be to score each of those axes on a 1 to 5 scale and then squint at the ones with the highest scores as your potential candidates.

那么,这个冲刺的第四部分,也就是最后一部分,实际上是可以一遍又一遍重复的。

Now, the 4th and final part of the sprint is actually one that you can repeat over and over again.

这个想法是,先推出一个共享agent,在你们已经能用的工具上去使用它,加载你们团队的上下文,然后至少两个人用它来做真实的工作。

The idea is to ship and start to use one shared agent on whatever tools you already have access to, loaded with your team context, used by at least 2 of you on real work.

当然,如果你们能用Claude Tag,那测试起来就很简单了,但我们也会有一些其他方法,让你们在这个体系内同样可以设置一个共享agent。

Now, certainly if you have access to Claude Tag, that's a really easy way to test this, but we'll have some ideas for other ways that you can set up a shared agent as well inside the system.

那么目标就是测试一个共享agent,给它一些时间,看看它是否以及如何改变工作方式,然后问问自己,它是不是真的让事情变得更好了。

The goal then is to test one shared agent, give it some amount of time to see if and how it changes the work, and then ask if it actually improved things.

有些用例,尽管你和其他人之间有很多共享的重叠,但目前可能并不适合让agent从中获得很多价值。

It may be that some use cases, despite there being a lot of shared overlap between you and others, just aren't the right shape for an agent to get a lot of value currently.

那没关系。

That's okay.

那就意味着你继续看下一个共享用例,看看agent在那里能不能帮上忙。

That means you move on to the next shared use case to see if an agent can help there.

同样,理论上,如果你愿意,只跑完这一轮就可以结束了。

And again, theoretically, after a single rotation of this, you can be done if you'd like.

但你基本上也可以反复运行这个流程,把你发现的每一个可能适合共享多人agent的潜在领域都过一遍。

But you can also run this basically repeatedly, working through each potential space that you uncovered that could be a good fit for a shared multiplayer agent.

所以这就是我们的想法。

So that's the idea.

就像我们所有的项目一样,这只是一个松散的框架,让你能够一头扎进去,通过实验来发现一些新的东西。

As with all of our programs, this is just meant to provide a loose framework that allows you to dive in and figure something new out through experimentation.
M1
M124:26

总的来说,对我来说重要的是,当我眯着眼看的时候,很明显这就是事情发展的方向。

Overall, the important thing to me here is that it seems fairly obvious to me when I squint at it that this is the direction things are headed.

再说一遍,不是说你们就不会用单独的agent了。

Again, it's not that you won't use individual agents.

你们绝对会用。

You absolutely will.

而且不只是用一两个,你可能会拥有整支agent舰队和团队要管理。

And not just one or two, but you will probably have entire fleets and teams that you manage.

但我确信的是,那只是我们工作的一部分。

What I'm convinced of though is that that only represents one part of the work that we do.

所以,我们开始为另一部分工作寻找agent也就很自然了,这部分工作就是人与人之间协作的那部分工作。

And so it is only natural that we start to figure out agents for the other part of work, which is the work that lives between us.

我觉得如果你和你的团队按照这个冲刺流程走一遍,当这些新的多人AI工具上线时,你们会处于一个明显更有利的位置,并且在这个令人兴奋的团队和多人AI新世界里成为领导者。

I think that if you and your team work your way through this sprint, you will be in a significantly better position to use these new multiplayer AI tools as they come online and be a leader in this exciting new world of team and multiplayer AI.

再说一遍,你可以在aidailybrief.ai找到所有这些链接,或者直接访问multiplayerai.ai。

Again, you can find all these links at aidailybrief.ai or go direct to multiplayerai.ai.

是的,那是连续两个AI,我很兴奋能和大家一起加入这个实验。

Yes, that is 2 AIs in a row, and I'm excited to join this experiment with all of you.

我们还会在AIDB Operator Circle上为此建立一个社区,所以也可以在节目说明里找那个链接。

We'll also set up a community for this on the AIDB Operator Circle, so look for a link for that in the show notes as well.

好了各位,今天的AI Daily Brief就到这里。

All right guys, for now that is gonna do it for today's AI Daily Brief.

一如既往地感谢你们的收听或收看,下次再见,peace。

Appreciate you listening or watching as always, and until next time, peace.
已剔除 5 处广告(点击展开查看)
M1
M10:52广告 · 已剔除

First of all, thank you to today's sponsors, KPMG.

Blitzy, Harbor, and HyperAgent.

To get an ad-free version of the show, go to patreon.com/AIDailyBrief, or you can subscribe on Apple Podcasts.

And to learn more about sponsoring the show, send us a note at [email protected].

Finally, last call, and blissfully for those of you who are not signing up, the last time that I will be yapping at you for a while about our Superintelligent Agent Executive programs.

The next cohort starts this week, so one last time, you can check them out at training.besuper.ai.

M1
M111:53广告 · 已剔除

Learn more about what separates AI Amplifiers from everyone else at kpmg.com/us/aiamplifiers.

Every AI coding tool on the market does the same thing first.

It starts writing code.

Blitzy does the opposite.

Before writing a single line, Blitzy spends days reverse engineering your entire codebase.

Thousands of agents ingest millions of lines, mapping every dependency, every undocumented constraint, every architectural decision made over the last decade.

The result is a dynamic knowledge graph that understands your software the way a principal engineer would after 30 years in the building.

Other tools guess at context with grep searches and markdown files.

Blitzy never guesses.

It builds true understanding first, then delivers over 80% of entire software epics autonomously.

Validated, end-to-end tested, production-grade pull requests.

That's why Fortune 500 engineering teams trust Blitzy with the codebases that matter most.

See for yourself at blitzy.com.

That's B-L-I-T-Z-Y dot com.

M1
M113:02广告 · 已剔除

Harbor Capital Advisors' AI Lab Ecosystem ETF suite gives investors a way to gain exposure to the AI ecosystem they believe is best positioned for success.

Search Harbor AI Lab Ecosystem ETFs wherever you invest or follow @HarborCapital on X to learn more.

Visit harborcapital.com for a prospectus containing investment objectives, risks, fees, expenses, and other important information.

Read and consider it carefully before investing.

Risks include principal loss and artificial intelligence-related risks.

Harbor ETFs are distributed by Foresight Fund Services LLC.

Harbor is not affiliated with AI Daily Brief, and the funds are not affiliated with, sponsored by, or endorsed by any AI lab.

This is a paid advertisement and not personalized investment advice.

Investing involves risk, including possible loss of principal.

This episode of the AI Daily Brief is brought to you by HyperAgent, where you run fleets of agents your team can manage together.

Forget local agents and chat workflows waiting on your laptop to be prompted.

HyperAgent deploys always-on agents in the cloud, doing real work across the tools your team already uses.

Marketing agents turn competitor moves into landing pages.

Sales agents enrich leads, draft emails, and updates the CRM.

Ops agent chases the paperwork and tracks the budget.

Every agent has access to shared context and follows your rules about scope and approvals.

It's time you had agents that feel like teammates.

Hire yours at HyperAgent.

Get $100 in credits at hyperagent.com/AIDailyBrief.

M1
M115:35广告 · 已剔除

When you want to collaborate with your teammates and agents, the best you can do is send a link to a read-only transcript they can't touch.

That's about to change.

Agents are starting to run tasks that take hours, days, even weeks.

Work at that scale was never meant to be done alone, and pulls in many people across a company.

Anyone on a team should be able to drop into the same live agent session to watch it work, redirect it, and hand it off.

The way they'd work with any other human team member.

This turns the work a team does with agents into a shared, living thing instead of 1,000 private threads.

We think there's a version of this for every kind of work: shared agents for engineers coding together in real time, for sales teams working a deal together, for support teams resolving a ticket, for lawyers drafting a contract, analysts building a model, and marketers shipping a campaign.

Anywhere a team already crowds around one problem, there should be multiplayer agents they all share.

Now we here at the AI Daily Brief are not addressing the startup and build side of that question, i.e., we're not releasing multiplayer agents for some specific use case.

But what I think that we can do is help your team build the foundations for using multiplayer agents and AI well.

Enter this fall's free self-directed education program, the Multiplayer AI Sprint for Teams.

You can find this linked from either the aidailybrief.ai website or the extremely clever direct URL, multiplayerai.ai.

The idea of the program is to build a shared 4-week sprint for your team to get yourselves ready to use multiplayer agents and then to actually try one in practice.

Now, at the time I'm recording, we're still finalizing some pieces of this platform.

So what you see on your screen now might change slightly by the time you're actually signing in.

But what I wanted to do was go through the 4 weeks at a high level to give you a sense of what the sprint involves.

Week 1 or session 1 of 4.

You certainly don't have to do it in just 1 week.

You can do it all in 1 fell swoop if you want.

But in any case, the first session is an inventory to find out what your team is actually running.

The goal of this is to take stock of how people on your team are currently using AI and agents.

Is everyone still just prompting ChatGPT or Claude?

Has anyone actually built or managed some agent that does recurring work for them?

Does anyone have context files or has even broached into using skills with an agent?

Almost inevitably on any team, there is going to be a wide range of what people are doing currently, but figuring out where everyone is is a really important first step in making the leap from single-player AI and agents to this multiplayer team type of usage.

As with each of the weeks, there will be 2 parts of this experience, a part that people do on their own, and then a meeting where they come together and share.

The part that you do on your own is to actually build the portfolio that helps you explain what you've done to your team.

Now, certainly you can do that on your own, but you can also use the builder that we have natively on the Multiplayer AI site to take a short interview with our AI to help you build that shareable portfolio automatically.

If you do choose to use our tool, it can contribute to a shared space that is just for you and the rest of the members of your team, where you can see all of the answers to how people are using AI altogether in one single spot.

M1
M123:29广告 · 已剔除

When you sign up, you'll get an invite code that you can share with other members of your team to make sure that you're all working in the same shared space together.

And if you choose to share your artifacts, can all resolve to the same spot.

Now, if you'll permit me, I would also be remiss at this point to not note that if you are sitting here feeling a bit overwhelmed and like you don't exactly have the foundations even on single-player AI to really be effective with this, I would recommend either A, going back through some of our free programs like the AIDB AgentOS program, Or if you want even more support than that, checking out some of our paid executive programs, specifically the Executive Catch-Up Program and the Executive Agent Leadership Program.

In particular, the Agent Leadership Program is going to leave you in the top 1% of people who actually understand how to build and interact with agents, which obviously gives you a huge advantage as you shift from thinking about single-player agents to multiplayer agents.

And the next iteration of that program is actually starting this week, just following Labor Day.

There's a link, of course, to that in the show notes.