我们如何打造“克劳迪”——AI项目经理（完整流程解析）

为了让人工智能对公司有用，需要进行协调一致的努力。

For AI to be useful to a company, it needs to be a coordinated effort.

克劳迪娅是我们与客户合作的咨询项目的项目经理。

Claudia is our project manager for the consulting work that we do with our clients.

他刚刚启动了四个子代理，来查看你的Gmail、日历、网盘和会议记录，以获取项目相关的联系人信息。

He just launched four sub agents to like look through your Gmail, look through your calendar, look through your drive, look through your meetings to like get contacts on the project.

然后它会收集这些信息，并将它们整理到你用于管理业务的电子表格中的正确位置。

And then it's gonna go and gather that information and then put it in the right place into the spreadsheets that you use to run the business.

我想暂停一下，说这太疯狂了。

Just wanna pause and be like that is that's crazy.

这确实有点疯狂。

That's kinda crazy.

我的意思是，比克劳迪娅做这件事更疯狂的是，替代方案是我来做这件事。

I mean, the only thing that's crazier is that the alternative to Claudie doing this is me doing this.

是的。

Yes.

我现在已经是个名副其实的氛围代码成瘾者了。

I am a bonafide vibe code addict at this point.

娜塔莉亚，欢迎来到节目。

Natalia, welcome to the show.

嘿，丹。

Hey, Dan.

很高兴见到你。

Good to see you.

很高兴能来这里。

Happy to be here.

我也很高兴见到你。

Good to see you too.

对于不了解的人，你是Every公司的咨询主管。

So for people who don't know, you are the head of consulting at Every.

我们认识已经很久了。

We've known each other for a really long time.

你虽然已经担任咨询主管一段时间了，但这次还是第一次上我们的播客。

Never been on the podcast even though you've been head of consulting for a while now.

我觉得你加入我们大概有九个月左右了，表现非常出色。

I think you've been with us for, like, nine months or so, and you've done a fantastic job.

所以我特别兴奋能邀请你来播客，和大家分享你的背景和专业知识。

So I'm just really excited to get you on the podcast and share who you are and what you know with people.

谢谢。

Thanks.

是的。

Yeah.

我非常热爱我们的社区和这个播客，很期待和大家聊天，也想听听其他人是如何看待咨询以及AI在他们公司中的应用的。

I I love our community and the podcast and just excited to chat and also hear how other people are thinking about consulting, you know, and and AI in their companies.

所以，是的，很高兴能来这里。

And so, yeah, happy to be here.

太棒了。

Awesome.

所以我觉得对你来说，分享过去九个月里你与全球一些顶尖公司交流如何进行AI部署的经验会非常有帮助。

So one of the things I think would be super helpful for you to share is over the last nine months, you've had a front row seat talk to some of the top companies in the world about how to do how they do AI deployments.

这些人都只是主动联系过来聊天的。

And that those are people that have reached out just to chat.

他们是你合作的客户。

Those are clients that you work with.

你为对冲基金、私募股权基金、财富500强企业以及许多你熟知的知名品牌做了大量的培训、集成和实施工作。

Do a lot of training and integration and implementation work with, like, hedge funds, PE funds, Fortune 500 companies, lots of main brands that you know about.

所以我觉得你一直坐在前排，亲眼见证了哪些方法有效、哪些无效，以及那些在AI采用和AI转型方面表现卓越的公司遵循了哪些模式？

And so I just feel like you've had this front row seat for, what works and what doesn't, and, like, what are the patterns that the companies that are really doing well at AI adoption and AI transformation are following?

我非常希望你能分享一些这些经验。

And I'd love for you to share some of those things.

是的。

Yeah.

确实如此。

That's true.

我的意思是，我们在Every所做的咨询工作让我们处于一个非常独特的位置，因为我个人在过去一年里与超过100家公司交流过，了解他们对如何使用AI的担忧，试图对标竞争对手是如何使用AI的，并且弄清楚哪些方法真正有效。

I mean, think we have been a real in a really unique position in the consulting work that we do at Every because, I mean, I personally have spoken to over a 100 companies in the past year, hearing their concerns around how they could be using AI, trying to benchmark how other competitors might be using AI, and then trying to get a sense of what actually works.

这归结为两件事。

And it really comes down to two things.

我们几个月前发过一篇帖子，讨论了从这些与我们交流过的约100家公司中学到的经验。

We talked about this in a post that we did a few months ago about kind of learnings from those 100 companies that we we talked to or so.

第一，要想在公司里有效使用AI，必须有一个有组织的行动。

And one is you really need an organized effort when it comes to using AI well in a company.

为了让AI对公司真正有用，它必须是一个协调一致的努力。

For AI to be useful to a company, it needs to be a coordinated effort.

为了让AI成为任何公司中高杠杆的工具，它必须自上而下地推动。

For AI to be a high leverage tool at any given company, it needs to come from the top down.

与过去那种软件应用不同——比如有人听说Asana对公司有帮助，就让CTO去买回来，然后指望员工会用。

So unlike historic, you know, kind of like software where, you know, someone heard that Asana was helpful for a company to use, then they just let the CTO sort of, buy it and then hope that people would use it.

如果缺乏协调一致的努力去理解公司内AI的潜在可能性、创造量身定制的机会以真正从中获取价值、跟踪员工的实际使用情况，并切实推行那些真正奏效的方法，那么AI根本就无从发挥任何作用。

If there isn't a coordinated effort to, understand what the possibilities are in using AI at a company, creating tailored opportunities to actually getting leverage and value out of those sort of use cases, and then tracking how people are actually using it, and then really implementing the the ways in which it works really well, AI really kind of doesn't go nowhere.

结果就是，只有少数几位高级用户能从中获得巨大收益，而其他人则茫然无措。

It ends up being that there are, like, a few high powered users that get a ton of leverage out of it, and then everyone else is sort of of floundering.

因此，我们在公司中看到有两个因素特别有效。

And so there's really two things that we see working well at companies.

第一，这是自上而下的推动。

One is it comes from the top down.

管理层认识到，AI是一项极具杠杆效应的工具，它从根本上改变了我们思考和对待工作的方式。

So leadership understands that this is a really high leverage tool, and it's fundamentally changing the way that we think and relate to work.

第二，他们为员工提供了成为AI倡导者和主人翁的机会，赋予他们探索如何重新定义自身角色、培训其他同事有效使用AI的权力，以应对我们当前所处的这一全新范式。

And two, they're really giving people an opportunity to become champions and owners of what it means to work with AI and creative power to explore how to rethink their roles and how to train other peers and other people to use AI really effectively given kind of this new, you know, this new paradigm that we're in.

因此，自上而下地推进，需要有协调一致的努力，让AI倡导者真正获得授权，去创造性地思考、尝试、实验，甚至在AI项目中失败，然后全力聚焦于真正有效的做法。

So coming from the top down, there's a coordinated effort and people, AI champions, really being empowered to think creatively, try, experiment, fail around AI initiatives, and then really doubling down on the things that really work.

是的。

Yeah.

这很有道理。

That makes a lot of sense.

我认为我可以从我的角度提供一些补充信息，因为我看到的没有你那么多，但我确实看到了很多。

I think some some color I can give from from my seat, because I I don't see nearly as much as you, but I do see a lot.

在自上而下的方面，那些不仅让我去做这件事、而且自己也亲身实践的CEO，才是推动公司走得最远的。

On the top down front, the CEOs that are actually doing not just sending me to do this, but actually doing it, like, those are the companies that go the furthest.

你在AI采用上能走多远，很大程度上取决于你的CEO走得多远。

You you will probably go as far as in terms of AI adoption, you'll probably go as far as your CEO has gone.

这并不是说CEO可以把事情委托给别人。

It's not saying that the CEO can delegate.

那些真正领先的CEO，他们自己就在用ChatGPT。

The ones that are really far ahead, it's like they're they're in ChatGPT.

他们也在用Cloud Code。

They're in Cloud Code.

他们不断尝试新东西，并对此充满热情。

They're, like, trying new stuff and being excited about it.

比如Shopify的Tobi，就是一个很好的公开例子，他周末就在自己动手搞这些玩意。

Like, you know, Atobi Atobi from Shopify is, like, a a good public example where he's just, like, hacking on stuff on the weekends.

你不一定非得做到那种程度，但Shopify的文化因为这一点，一年后会大不相同，我觉得这非常重要。

You don't necessarily have to be that far, but, Shopify, its culture will be is a lot different and will be way different in a year because of that, and I think that's really important.

所以这就是自上而下的方式。

So that's the sort of top down.

然后我认为你所暗示的自下而上的方式是，在任何组织内部，都有一些天生的早期采用者，而你作为领导者的工作就是识别这些人，传播他们的知识，提升他们的地位，让他们能为那些可能非常有价值但不会主动尝试新技术的人铺平道路。

And then I think the bottoms up thing that you're you're alluding to is inside of any organization, there are people who are just natural early adopters, and your job as an executive who's leading your org is to identify those people and spread what they know and elevate their status so that they can kind of pave the path for for everyone else who is maybe, like, super valuable, but isn't is not naturally just gonna, like, try some new technology.

但他们如果被展示出，嘿，

But but will use it if they're shown, hey.

这确实能帮助你更好地完成工作，就会去使用它。

This is, like, actually something that is gonna help you in your job.

是的。

Yeah.

我的意思是，Dan，我经常有这种感觉。

I mean, Dan, I feel this with you all the time.

你知道的？

You know?

新模型会推出，你会问：为什么你没用这个模型测试过？

A new model will come out, and you'll be like, why haven't you run this through x model?

我会说：你说得对。

And I'll be like, you're right.

我为什么没用这个模型测试呢？

Why haven't I run it through this model?

但你知道，我在内部也看到，它在每周一的站会上自然地浮现出来——有人会说他们在尝试一个全新的使用场景或应用，然后我们其他人就会意识到，这带来了新的可能性，令人兴奋。

But, also, you know, I see it internally in the ways that it comes up naturally when in our Monday stand ups, someone will say they were tinkering with a whole new use case or application, And then the rest of us will sort of see there's this new dimension of what is possible, and it's exciting.

当它奏效时，真的很令人兴奋，但你需要处于一种敢于尝试、敢于失败的创意环境中。

When it works, it's exciting, but you need to be in this creative space where you're trying, you're failing.

它其实并不奏效。

It's not really working.

它真的奏效了。

It is really working.

它非常强大。

It's really powerful.

强大。

Powerful.

当你看到可能实现的事情时，你才会真正明白，你能走到多远，它能带你走多远。

And then when you see what's possible, then you really understand, like, where you can go and just how far it can take you.

你有没有一些具体的故事？当然，我们不能透露客户的姓名，但你有没有见过一些特别有力的突破案例？

Do you have any specific stories, you know, obviously we can't share from clients, like, by name, but do you have any specific stories about unlocks that you've seen that were particularly powerful?

也许是我们和客户合作中的一些案例，或者你看到的一些可能看起来有点反直觉的事情，因为我想听的人可能会想，是的，这听起来不错。

Well, maybe it's something that we've done with our clients or just something that you've seen that maybe feels a little bit counterintuitive because I can imagine people listening to this and being like, yeah, that sounds good.

一般来说，让CEO们接触AI确实很棒。

Like, generally, it's great at CEOs into AI.

一般来说，内部推广核心用户确实很好，但我想深入探讨一些具体的、反直觉的、或者投入很少但回报巨大的实际案例，你觉得有哪些？

Generally, it's great to to promote your power users internally, but, like, wanna get down into the nitty gritty of here's some, like, actual concrete stuff that is maybe a little bit counterintuitive or is a really big unlock versus the effort required?

你想到什么了吗？

Like, what comes to mind?

我想到两件事。

There's there's two things that come to mind.

一个是去年开始合作的一家私募股权公司，我们至今仍在合作。

One is with a private equity firm that we've we've started working with last year that we're still working with.

这个例子之所以让我想到，是因为我们在这家公司日常对接的合作伙伴，不仅技术能力出众，他的真正优势在于对AI相关的人际动态有着非常深刻的理解。

And that one comes to mind because our partner, kind of like our day to day liaison at that firm, is both brilliant technically, but his superpower is actually that he understands the people dynamics around AI really, really well.

由于他主动承担了在公司内推广AI的职责。

And because his role is he's taken it upon himself to roll out AI at his firm.

他明白这不仅是技术挑战，更关键的是人的问题。

And he understands that it's a technical challenge, but what he really understands is that it's a people challenge.

由于他所在的是一家私募股权公司，和其他许多投资公司一样，人手和时间都十分有限。

And because he's at a private equity firm and like a lot of other investment firms, there's a lack of bandwidth.

大家没有足够的余力去尝试新技术，或者承受失败的风险。

There's a lack of capacity to, you know, try new technology, see it fail.

有些团队已经比较先进，正在以非常有趣且高级的方式使用AI。

There's people that are more advanced, teams that are already using it in pretty advanced kind of interesting ways.

而其他一些团队则一直缺乏实施AI的资源和空间。

There's other teams that just haven't had capacity to implement it.

因此，我们共同做的一件事是，我们一开始的合作就是他坐下来与他公司的投资者们沟通。

And so one of the things that we did together is we started out our work together by he sat down with the investors at his firm.

我认为，他基本上把他们所做的每一项任务都详细地梳理了一遍，从研究、尽职调查、市场分析到投资组合管理，一直到投资者日常生活的方方面面，事无巨细。

And I think, you know, he basically mapped out every single task that they do in, like, to the most detailed kind of like end, every single task that they do from research to diligence to market mapping to portfolio management to just kind of, like, the day to day of, like, running their lives as investors.

最终，我们得到了一个极其详细的视角，展现了这家公司的投资者是如何完成工作的，同时也按团队进行了区分，因为不同策略下的工作方式差异很大。

And so what we ended up with was the starting point of a very, very detailed view of what it looks like at this firm for an investor to do their job and what that looks like also by team because it can vary quite a lot depending on what the strategy is.

然后，我们审视了这份冗长的任务清单，逐一标记出那些能够带来极高杠杆效应的AI应用机会。

And then what we did is we looked at that long task, that long list of tasks for that firm, and we went through and highlighted where there are opportunities to use AI that are really high leverage.

因此，我们最终为所有合作客户创建了这样一张地图。

And so what we ended up with is this is this this map that we end up creating for all of the clients that we work with.

但由于它如此详尽，我们能够极其精准地寻找解决方案，不仅为团队释放人力，更在我们共同开展的培训和工作中实现真正的高杠杆效果。

But it was so detailed that we could really be very, very precise about looking for solutions that would give the team not just bandwidth, but really high leverage in any of the training and work that we did together.

只有当有内部人员不仅描述团队普遍依赖的工作流程，而是极其具体地阐述他们自身的工作方式和思维模式时，这种工作才有可能实现。

And that that's the kind of work that's only possible when you have someone on the inside who is not just describing the work and workflows that teams rely on generally, but, like, very, very specifically to the work that they do and the the way in which they approach or think about their work.

这使得我们在共同开发的培训、赋能和工具中，出现了一个令人惊叹的时刻——投资者们突然意识到，他们现在可以用短短三十分钟，就写出一份过去需要两到三周才能完成的高质量投资备忘录。

And it's made it so that in the training and the enablement and the tools that we've been able to develop together, there's this moment, this, like, moment that is, like, wild, you know, where investors will kind of come and realize that there is a new way that they can write an investment memo that previously took two to three weeks, and they can now get a really high quality draft in literally thirty minutes.

只有当你有一个内部人员，了解所有细节时，这才有可能实现。

And that's only possible when you have someone on the inside who understands all of the elements.

这很有趣。

That's interesting.

能再详细说说那个时刻吗？那种自动化或工作流程具体是怎样的？

Tell me more about that moment and, like, what that kind of either automation or workflow, what what is that?

他们是很好地使用了ChatGPT吗？

Is that, like, they're using ChatGPT well?

他们是很好地使用了CloudCode吗？

They're using CloudCode well?

是的。

Like, yeah.

深入讲讲这个。

Go get more deep into that.

在这种情况下，涉及几方面。

So in that case, it's a few things.

一是这家机构在SharePoint上积累了大量关于其投资理念的资源。

One is this particular firm has a lot of resources in SharePoint around the theses that they have, their investment theses.

而这正是该机构的知识产权。

And this is this is kind of like the IP of the firm.

对吧？

Right?

他们花了十年甚至更长时间，深入思考如何应对特定的投资领域。

They've spent a decade, if not more, really thinking about how they approach a particular area of investment.

当他们对一家新公司进行尽职调查时，他们希望基于这十年积累的知识库，思考如何从超越数字计算的角度来评估这个机会。

And when they're diligencing a new company, they wanna understand, given this repository of knowledge that they've accumulated over a decade, how should they be thinking about this opportunity, you know, beyond just the number crunching?

而这确实是一项非常繁重的任务。

And that's something that is, like, really quite onerous.

要真正阅读并消化这些内容，再将其进行对比，是一项巨大的工作。

It's a huge task to take on to, like, really read that and then digest how it compares.

当然，这一点正是ChatGPT能够非常有效地完成的。

And, of course, that's something that just CHAP GPT is able to do effectively very, very effectively.

对他们来说，这看起来像是连接正确的上下文，也就是合适的数据来源，然后通过一个经过训练的提示词来理解他们对特定投资策略的思考方式。

And so what it looks like for them is connecting the right context, really, the right sort of sources of data, and then funneling it through this prompt that is trained to understand how they think about that particular investment strategy.

然后，是的，基本上就是创建一组 GBT 和提示词，使得将所有这些信息综合成一份投资备忘录变得非常容易，这份备忘录能为他们提供一个总体框架，帮助他们评估这家公司与更广泛机会以及他们十年积累的信息之间的对比关系。

And then, yeah, basically, just creating kind of, like, a a set of, like, GBTs and prompts that make it that it's really easy to synthesize all of that information into an investment memo that really gives them sort of that general rubric of how that company compares to this broader opportunity and to the decade of information that they've collected.

这原本是分析师和助理合伙人需要花两到三周时间才能整理出来的内容，然后才提交给投资委员会。

That's something that, again, an analyst and associate principal spend two to three weeks to pull together before it goes to the IC, the investment committee.

而现在，你可以在三十分钟内得到一份非常扎实的初稿。

And so now you get a really solid draft in thirty minutes.

这真的很有趣。

That's really interesting.

我认为这实际上是我们看到的许多公司中的一种普遍模式。

And I I think that's, like, actually a broad general pattern that we see in a lot of companies.

甚至在我们自己的公司里，最明显的第一步就是把 AI 连接到所有的数据源上，这虽然很难，但这是基本要求——它必须连接到所有上下文所在的地方。

Even in our own company is, like, the first one is the obvious one is the connect the AI to all your data sources, which is it's hard, but, like, that's that's sort of table stakes is it needs to be connected to the place where all the context lives.

但另一件正在发生的事情是，尤其是随着我们的团队以及许多其他团队逐渐转向更偏向于智能体的模式，比如使用 Cloud Code、Cowork 或其他各种工具，人们期望智能体能独立去完成一些工作，持续二十分钟，而不再像以前那样只是进行来回的对话。

But then the other thing that's been happening, especially as, you know, our org and lots of other orgs that were around are transitioning more into a, like, agent, let's say agent native world where they're using Cloud Code or they're using Cowork or they're using, you know, all these other kinds of tools where you kind of expect the agent to be going off and doing some work for twenty minutes, and it's not necessarily like a back and forth chat in the same way.

一旦你连接了所有数据，就非常关键的是，你所构建的提示词或技能要能告诉AI：嘿，这里是你要找特定信息的方法。

Once you have the connections to all the data, it's really important to have the prompt or the skill that you've built be able to tell the AI, like, hey, here's how you find the specific thing that you want.

比如，对我们来说，如果你想了解我们的收入情况，实际上有三个不同的地方可以查。

Like, for example, for us, if you wanted to figure out what our, like, what our revenue is, like, there's like three different places you could go.

你可以去Stripe，也可以去Chart Mobile，也许还可以去Posthog，但我们的增长负责人奥斯汀对我们的MRR有一个特定的定义方式。与其让AI每次都从零开始推断，不如直接明确告诉我们：这就是我们对MRR的理解。这种思路也可以迁移到为某个客户做咨询时：这就是我们对这个行业的看法，以及你该从哪里获取这个特定行业的数据。

You can go into Stripe, you can go into Chart Mobile, maybe you can go into Posthog, but like, our head of growth, Austin, has like a particular, like, way that he's defined what our MRR is that instead of forcing the agent to, like, figure that out from scratch every single time, sort of putting into place, here's how we think about what MRR is, and that transfers into, like, know, for consulting for for one of these clients, like, here's how we think about this sector and here's where you get the data for this particular sector.

这正是价值的核心所在，也是你的AI使用方式区别于他人使用方式的关键。

That's, like, where a lot of the value is and and a lot of what makes your use of AI different from someone else's use of AI.

是的。

Yeah.

完全正确。

That's totally right.

我认为这其实是AI最难的部分，而我们正在合作的这家机构之所以如此神奇，就在于我们的合伙人乔纳森——他逐一访谈了每一位投资者和每个团队，深入理解团队在撰写投资备忘录的每个环节中集体思考的细微差别。

I I think it's the hardest part of AI, actually, and this is the part that has been so magical at this particular firm that we've been working with, is that our our partner, his name's his name's Jonathan, Jonathan basically interviewed every single investor and every single team to really understand the the the nuance in which a team collectively thinks about every part of the investment memo.

如果没有如此高度的定制化，我们今天所完成的这项工作根本不可能实现。

And this work that we've been able to do together really would not have been possible if it didn't have such a high degree of tailoring.

你知道吗，这就像西尔维·罗那样的提示定制。

You know, this is like this is like Savile Row sort of prompt tail tailoring.

它非常、非常、非常具体。

Like, it's so, so, so specific.

数字的呈现方式、图表的展示方式，以及他们内部对这些内容的表达和思考方式，都非常重要。

The way that numbers show up, the way that figures show up, the way that they express or think internally around this stuff, it's really important.

而这些提示正是反映了这一点。

And the prompts reflect that.

因此，这些提示最终变成了能够完成高质量、可靠工作的分析师，这太棒了。

And so the prompts really end up being this analyst that does really high quality work that is dependable, and that's so cool.

这真的很酷。

That's really cool.

我知道我们还与对冲基金和科技公司做了大量工作。

I I know we're we've also done a lot of work with hedge funds and also with tech companies.

在这些领域里，你还有其他想分享的例子吗？

Any other examples you wanna share in those domains?

是的。

Yeah.

我的意思是，让我想想。

I mean, let me think.

有太多酷的应用了。

There's so there's so many cool applications.

也许我可以谈谈我们目前正在合作的一家科技公司中观察到的一个非常有趣的模式。

Maybe I'll speak to there's a really interesting pattern that we're that we're seeing at one of the tech companies that we're working with right now.

我们知道，在与工程师和工程团队合作时，实施人工智能有一个四步流程效果很好：先规划，再委派，然后评估，最后巩固有效做法或从该次实践中获得的经验。

We know that when it comes to working with engineers and with engineering orgs, there's sort of like a four step process that works well when it comes to implementing AI, and that is you plan, you delegate, you assess, and then you compound what works or kind of like the learnings of that that particular session.

当我们与这家公司的工程师交流时，发现他们在委派、评估甚至经验积累方面都非常有效。

And when we spoke to the engineers in this particular org, we found that they were actually really effective at the delegating, at the assessment, and even the compounding.

但他们完全没有规划阶段，因此进展有限。

But there was no planning phase, and so they weren't going very far.

他们不断遇到同样的问题，因为缺乏一个良好的计划来搭建有效的框架，使他们能够解决许多小问题，却无法应对那些我们期望AI帮助解决的更重大、更核心的挑战。

And they were running into the same sort of challenges over and over again because there wasn't a good plan for them to really scaffold significant workaround so they could solve a lot of small issues, but they weren't able to address these big sort of, like, meatier problems that we kind of hope for AI to help with.

因此，只有通过理解这群工程师是如何使用AI的，我们才能真正意识到——他们只是缺少了规划阶段。

And so this is the kind of thing that only by understanding how that particular, you know, group of engineers was using AI could we really realize, like, you're just missing the planning phase.

对吧？

Right?

我们需要做的，就是帮助他们理解什么是良好的规划。

Like, we just need to do enablement around what good planning looks like.

而且我们已经看到，正如我们都知道的，这带来了巨大的改变。

And we're already seeing that, as I think we all know, it makes a huge difference.

你所能积累的成果，完全取决于你的规划有多好。

And you can only really compound as much as you plan.

对吧？

Right?

现在，随着他们开始实施这些大型规划并推动重要工作的开展，我认为我们正开始看到那种在工程团队中期待的高杠杆效应。

So now that they're starting to compound these big plans that are developing significant work, I think we're starting to get that sort of high leverage machine that we hope to see work in engineering orgs.

你认为这里可能实现什么？

What do we think is possible here?

比如，我们期望看到哪些方面的效率提升？

Like, what are the kinds of speed ups that that we expect?

特别是工程领域？

In engineering in particular?

是的。

Yeah.

要知道，这个问题很难回答，但我想说的是，当这个计划-委托-评估-累积框架被正确实施和使用时，我们经常看到这样的情况。

Know, that's a difficult question to answer, but, you know, I would say we're consistently seeing when this plan delegate assess compound framework is in place and used well.

工程师们常常能在一下午的时间内高效完成原本需要两周的工作量。

We're frequently seeing engineers generate two weeks of work effectively in an afternoon.

我不惊讶于这种效率还会继续提升。

And I wouldn't be surprised if that continues to speed up.

对。

Yeah.

我的意思是，我们也有同样的观察，这确实改变了我们对技术团队招聘人选、优化目标，甚至编程面试和技术面试方式的思考。

I mean, we see that too, and it it it it definitely changes how we think about who we can hire on the technical side and what we are optimizing for and even how we do, like, you know, programming interviews and stuff like technical interviews.

这真的是一种非常有趣的改变，但我认为更有趣的一点是，对于你来说，我最近三四周一直看着你和组织里其他几位非技术人员，完全被震撼到了。

It's it's like a really in it's a really interesting change, but I think one of the more interesting ones is for you, I've just watched you and several other people who are not technical inside the org just get totally like, your your your mind totally blown over the last, like, three or four weeks.

感觉就像发生了一次巨大的范式转变，我以前只是发消息给你，你就会说，是的。

Like, it feels like there has been this, like, massive phase shift where I would just, like, message you, you'd be like, yeah.

我从早上六点就开始写代码了。

I've been up since, like, 6AM by by coding.

你能跟我们说说这件事吗？

Can you, like, can you tell us about that?

因为我觉得这非常有意思，而且如果要我猜的话，你可以说是走在最前沿的人，之后会有很多人跟上你，感受到同样的变化，我们会把你现在学到的这些经验推广给客户，推广给所有观看这类视频的人，因为这是一种非常有价值的新工作方式。

Because I think it's really interesting, and I think that you're if I had to guess, you're sort of the leading edge of the spear, and there's gonna be a lot of people coming after you that are feeling the same way, and that we're gonna spread a lot of things that you're learning right now to our clients, and just really anyone who's watching videos like these because it's a it's a new way of working that really valuable.

说实话。

So I'll be honest.

所以，是的，我得承认，我现在已经是个名副其实的Vibe编码成瘾者了。

I so, yes, I I I think I will admit that I am a bonafide vibe code addict at this point.

这件事的发生过程很有趣，说来挺有意思的。

So I the way this happened is a funny it's a funny thing.

我其实意识到，我正开始陷入一个我经常看到客户陷入的陷阱。

I I actually realized I was starting to fall into a trap that I often see our clients fall into.

所以去年年底，你知道，我们有太多项目在同时进行。

So at the end of last year, you know, we had so many projects going on.

我们正在为组织内数百人提供支持。

We were supporting hundreds of people across organizations.

每天我的一天都从一堆会议和一堆工作开始。

And every day, my day would start, and I just had a bunch of meetings and a bunch of work to do.

所以我根本没有时间去尝试这些工具。

And so I didn't really have time to play with a lot of these tools.

进入今年后，我们和尼特什·阿加瓦尔意识到这一点，他是我们咨询团队中的应用AI工程师，非常出色。

And going into this year, we realized with Nitesh Agarwal, who is our applied AI engineer that is in on the consulting team and is fantastic.

他之前是Cora的工程师，帮助打造了这个出色的产品，然后转入咨询部门，帮助放大我们为客户所做的工作。

He was previously an engineer at Cora and helped build this beautiful product and then moved into the consulting org to help, you know, amplify the work that we're doing with our clients.

我们和尼特什意识到，如果我们把工作安排在朝九晚五的框架内，就无法像我们希望的那样快速推进并完成有创意的工作。

We realized with then with Nitesh that we weren't going to move as fast and do the creative work that we wanted if we were scheduling the work to happen in the nine to five, if you will.

所以我们决定提前三个小时开始一天，早上六点就开会。

And so we decided to start our day three hours early so we would meet at 6AM.

我们基本上从六点到九点一起轻松地写代码。

And we would basically just vibe code from, like, six to 9AM.

这一切始于我们问自己：我们能否创建一个非常雄心勃勃的项目？因为对任何咨询公司来说，项目管理都极其耗时。

And it all started with us asking, you know, could we create this really ambitious project, which is, you know, project management is really time consuming for any consulting business.

对吧？

Right?

任何优秀的咨询公司都有一个专门的项目管理职能。

Any great consulting business has an entire function around project management.

这确实是一门技术，但也需要理解大量动态因素，比如客户偏好的信息呈现方式、他们如何安排会议，以及每个项目中那些细微的细节。

And it's a real skill, but it it also requires, you know, understanding a lot of moving pieces, how clients prefer information, how they are scheduling sessions, like, of the sort of, like, nuance things that are happening for any given project.

我和尼沙什于是问：我们是否真的能启动一个代理，雇一个代理来当我们的项目经理？

And Nishash and I sort of asked, you know, do we think we could really spin up a we could spin up basically like an agent, hire an agent to be our project manager.

答案很快就出来了：可以。

And the answer really quickly was yes.

但要有效实现这一点的框架——具体来说是在Claude代码中——实际上经历了大量迭代。

But, also, the the framework for how to do that effectively, this is in Claude code specifically, actually took a lot of iteration.

我想说，我们三次都达到了85%的进度，但后来因为学到的新东西不得不推倒重来，重新构建一个真正达到100%效果的新框架。

Like, I would say we got 85% of the way there three times and then had to scrap it given kind of what we learned and then start again to get to somewhere to get to a new framework that actually got us to a 100%.

所以这真的特别有趣。

So it's just been so fun.

真正构建出一些东西太酷了，我觉得这完全是富有创造力的工作。

It's so cool to really build something, and it's just I think it's really creative work.

对吧？

Right?

而且，深入思考这些问题，真正理解什么是优秀的项目管理者，也让人豁然开朗。

And it's really also clarifying work to really understand to think about the questions, what does it mean to be a good project manager?

对于我们在经营的业务来说，什么样的项目管理者才算优秀？

What does it mean to be a good project manager for me in the business that we're running?

我又该如何把这些经验转化为一系列明确的指令？

And how do I codify this into a series of instructions?

你知道，我们谈论人工智能、有效使用人工智能，其实很大程度上是关于成为一个优秀的管理者。

You know, we talk about AI, using AI effectively, being a lot about being a good manager.

关键在于，我该如何成为一个清晰的沟通者，提供明确的指令，以便我们能够真正创建这个名为Claudie的智能体，让它独立运行并为我们完成工作？

It's like, how do I how do I how can I be a clear communicator and provide clear instructions so that we can really create this agent, we call Claudie, to really run on their own and to do this work for us?

就在几周前，这个系统刚刚成型，现在看到它真正运行起来，真是太棒了。

And it's just so cool to be a few weeks out from that and to really have this system working.

增长是好事。

Growth is good.

速度是好事。

Speed is good.

但快速增长也会带来特定的代价。

But there's a specific tax that comes with growing fast.

每个活动都需要一个新的着陆页。

Every campaign needs a new landing page.

每个产品发布都需要更新文案，每个A/B测试都堆积在待办列表里。

Every product launch needs copy updates, and every AB test sits in a backlog somewhere.

你的网站成了拖慢你进度的瓶颈，而你根本没有足够的工程师来跟上需求。

Your site becomes the thing slowing you down, and you just don't have enough engineers to keep up.

Framer 是一个像你团队最爱的设计工具那样工作的网站构建器，支持实时协作、功能完备的 CMS（包含所有必要的 SEO 功能），以及包含集成 A/B 测试的高级分析，让你的设计师和市场人员从第一天起就能自主构建并优化你的 .com 网站。

Framer is a website builder that works like your team's favorite design tool With real time collaboration, a robust CMS with everything you need for great SEO, and advanced analytics that include integrated AB testing, your designers and marketers are empowered to build and maximize your .com from day one.

Perplexity、Miro 和 Every 等公司都在使用 Framer，因为它消除了瓶颈。

Companies like Perplexity, Miro, and Every use Framer because it eliminates the bottleneck.

市场团队可以测试创意，设计师可以发布更新，完全无需排队等待工程师处理。

Marketing can test ideas, designers can ship updates, all without touching the engineering queue.

更改将在几秒钟内上线。

Changes go live in seconds.

一键发布，搞定。

One click, publish, and done.

它专为与你共同成长而设计。

It's built to scale with you.

提供高级托管服务、99.99% 的正常运行时间服务等级协议、企业级安全防护，以及其他更多功能。

Premium hosting, 99.99% uptime SLAs, enterprise grade security, and more.

无论你是首次上线网站，还是迁移一个已有的.com网站，Framer都能让你的网站从负担转变为优势。

Whether you're launching your first site or migrating in an established .com, Framer turns your website from a constraint into an advantage.

了解如何通过Framer专家让你的.com网站发挥更大价值，或立即免费开始构建，访问 framer.com/dan，享受Framer Pro年费计划30%折扣。

Learn how you can get more out of your.com from a Framer specialist or get started building for free today at framer.com/dan for 30% off a Framer Pro annual plan.

访问 framer.com/dan，享30%折扣。

That's framer.com/dan for 30% off.

规则和限制可能适用。

Rules and restrictions may apply.

回到本集。

Back to the episode.

这真的很酷。

That's really cool.

我想让你给我们展示一下这个系统。

I want you to I want you to show us the system.

在我们进入那之前，我想指出一个有趣的现象：你并没有只期待它在朝九晚五的工作时间内发生，而是特意从工作之外抽出三小时来玩。

Before we get there, I wanna point out the that interesting pattern, which is instead of just expecting it to happen inside your nine to five, like, actually just carved out three hours outside of your job to, like, play.

我认为这是一个很有趣的教训，我们确实在Every内部深深体会到了这一点。

And I think that's, like, a interesting lesson that we definitely just know ourselves internally inside of Every.

我们刚从思想周回来，每六个月我们都会在巴拿马聚在一起。

Like, we just got back from Think Week, which we do every every six months where we're all in Panama together.

我们彻底放下了日常的工作，目的就是自由地探索技术、互相了解、创造有趣的东西。

And we just got rid of all of our day to day work, and the whole point was just to, like, play around with your technology, get to know each other, build interesting things.

就是单纯地玩。

Just just play.

想做什么就做什么。

Like, do whatever you want.

我认为这在技术飞速变化的时代尤为重要，因为你不想拼命成为跑得最快的马车司机，你知道吗？你必须先从马车赛中抽身出来，好好看看这车到底是什么，才能学会开车。

And I think that's so important in a world where technology is changing so fast because you what you don't wanna do is, like, work really hard to be the fastest horse and buggy driver, you know, and it and you can't learn to drive a car until you, like, take some time out of your horse and buggy race to be like, what is this car thing?

你知道吧？

You know?

我觉得你正是发现了这一点。

And I think you you discovered that.

我认为这是我们公司在Every内部一直在做的事情，也是许多客户以及我认为做得很好的公司所熟知的：给予人们所需的空间，让他们能够以无压力的方式尝试新技术，而不用担心拖慢工作进度。

I think that's something that we've done inside of every and it's also something that a lot of our clients and just generally companies that I think do this well know how to do is give people the space that they need in order to feel like they can try out new technology in a rift free way, where they're not gonna get behind in their job.

他们可以慢慢了解其中的门道，甚至允许自己失败。

They can kinda like learn its ins and outs and fail.

经过几次这样的迭代后，他们会说：天啊。

And then after a couple times of the you know, iterations of this, they're like, holy shit.

我现在已经是在开车了。

I had this I'm driving a car now.

我不再是开马车和马车夫了。

Like, I'm not driving a Wolf and Buggy anymore.

我认为这非常有价值。

And I think that's so valuable.

但这真的很难。

And it's really hard.

实际上，拥有这种创造性的空间，与我们通常的工作方式是完全背道而驰的。

Actually, having that creative space is very, very counterintuitive to the way that we usually work.

对吧？

Right?

在传统工作中，我们实际上花了多少时间在摸索新的做事方法上呢？

How how much of our time is really spent, you know, in traditional jobs, just like figuring out, like seeing if there's a new way to do things.

通常，当你被雇佣去做一份工作时，雇主会给你一套明确的职责，让你一直做下去，直到你晋升到下一个级别，无论那是什么。

Usually when you're hired to do a job, you're hired to do a specific set of functions that have been laid out to you that you're supposed to do until you get to the next level, whatever that is.

而一家公司敢于说：‘我们认为这一切都在变化，我们还不确定具体怎么变，但我们相信你能自己找到答案，理解这对你的意义，并且我们可能会引入外部合作伙伴来加速这个过程’——这真的非常具有革命性。

And for a company to be so bold as to say, hey, we think this is all changing, and we don't know exactly how it's changing, but we trust that you can figure it out, and you can figure out what this means to you, and maybe we'll bring in outside partners to accelerate the way in which you do that, it's really revolutionary.

这真的太棒了。

It's really amazing.

但同时，这也需要一个允许失败、鼓励实验的创造性环境。

But also, it can be a really creative space where you have to be at a company that's willing to see things fail, to experiment.

对吧？

Right?

我们曾经三次丢弃了我们的项目管理代理，直到找到真正有效的框架，这每周为我们节省了大量时间。

Like, we had to throw our project manager agent away three times before we found the scaffolding that really works, and that saves us so much time per week.

但我不是工程师。

But I'm not an engineer.

我不是产品经理。

I'm not a product manager.

这并不是我日常的工作内容。

This isn't kind of like my day to day job.

只有当你戴上创意的帽子，不断尝试，直到找到真正有效的方法时，这一切才会变得令人兴奋且可能实现。

And this is only exciting and possible when you put on this creative hat and just keep on tinkering until you find something that really works.

对我来说，有Nitesh这样的伙伴，还有你和我身边每一位团队成员，我不断看到什么是可能的，这让所有这些事情都变得更容易实现。

And for me, having a Nitesh, you know, for me having these, like, incredible resources, you and everyone on the every team around me where I'm seeing constantly what's possible, it just makes all of these things so much more achievable.

是的。

Yeah.

我认为另一个很好的模式是，Nitesh是一位应用型AI工程师，他可以真正和你坐在一起，帮助你弄清楚：基于我的工作流程，我该如何构建这个项目管理工具？

I think that's another really good pattern is you have Nitesh who's an applied AI engineer who can literally sit with you and help you figure out, okay, given my workflow, how can I build this project manager?

你了解自己需要什么，而他则掌握着技术前沿的知识。

And you have the expertise in, like, what's needed, and he has the expertise in, like, what's at the edge of technology.

我认为，我看到很多CEO都在实践另一个很好的模式：公司的前进速度，取决于我自身对AI的掌握程度。

And I also I think that that's another really good pattern I see a lot of CEOs doing is, okay, the company's gonna only go as far as I go in terms of knowing how to use AI.

我会找一个真正懂AI的人，和我坐在一起，帮我理清思路：我脑子里有个项目，我觉得如果能完成，一定会非常有趣且有价值。

I'm gonna have someone who knows about what they're talking about in AI, like, just literally sit with me and talk me through, okay, I have this project on my mind that I feel like would be really fun and really valuable if I got it done.

这会是一半学习、一半实践的过程，我们一起努力实现这个充满野心又有趣的目标。

It's gonna be half learning, half us just, like, trying to knock out this, like, really ambitious, interesting thing.

我认为，这确实是让自己对AI上瘾的好方法：当你刚刚试探性地踏入这个领域时，有个人就在你身边陪着你。

And I think that's actually a really good way to get in get yourself addicted to vibrating is have someone who's sitting next to you as you put your toe in the water.

是的。

Yeah.

当然。

Absolutely.

我还要说，你需要的工程能力和AI知识，和你对‘什么是优秀成果’的理解一样重要，而这种理解因你的具体工作而异。

I'll also say I think there's something to be said about you need as much engineering power and, like, sort of like AI know how as you need an understanding of, like, what good looks like, which is very specific depending on what it is that you do.

我们之前那些不成功的智能体版本，其中一个就是过于偏向工程了。

And the different iterations of our agent that didn't work were one was just too engineering focused.

它太专注于框架以及数据之间如何连接的策略。

It was too focused on the framework and the strategy of how the data would be connected to each other.

另一个则过于专注于工作本身。

The other one was too focused on just what the work is.

对吧？

Right?

所以它就像一份职位描述。

So it was just kind of like a job description.

直到我们意识到，好吧。

And it wasn't until we realized, like, alright.

它需要结合对优秀项目管理的理解，以及对我们而言什么是好的——也就是我所掌握的信息和背景。

It's a mix of the know how of, like, what good project management looks like and what it looks like to us, which is kind of the information and context that I have.

同时，还要知道如何最好地组织任务、代理、子代理以及所有这些Claude代码基础设施，以满足我们具体想要解决的需求，这才真正融合并奏效了。

And then also, you know, how tasks and agents and sub agents and all of this Claude code infrastructure can best be organized so that it serves kind of the need that we're specifically looking to solve, that it really came together and worked.

所以你知道，即使你拥有强大的工程能力，你也需要具备实现有用成果的诀窍。

And so you you know, you can have really great engineering power, but you also need to have, like, the know how to get to something useful.

好的。

Alright.

听起来不错。

Sounds good.

你愿意给我们展示一下Claudie吗？

Do you wanna show us a little bit of Claudie?

是的。

Yeah.

我们开始吧。

Let's do it.

好的。

Okay.

太好了。

Cool.

你能看到我的屏幕吗？

Can you see my screen?

我能看到你的屏幕。

I can see your screen.

好的。

Alright.

欢迎来到Every Consulting的GitHub页面，这里就是Claudia的家。

So welcome to the every consulting GitHub page, and this is where this is where Claudia lives.

Claudia是我们为客户做咨询工作的项目经理。

So Claudia is our project manager for the consulting work that we do with our clients.

我先给你们展示一下架构，我觉得挺酷的。

And the first thing that I'll show you is the architecture, which I think is pretty cool.

这个架构花了大约两周时间反复打磨，才最终实现。

This this took, you know, two weeks to, like, really refine and come up with to have it work.

我其实不会深入讲解这些细节。

So I won't you know, I actually won't I won't go too into the details of this.

我们有一篇很棒的文章，稍后会分享，里面会详细说明这套架构的搭建方式。

We actually have a great post that we'll share and go into the details of how this is set up.

但在最高层级上，我们有一个Claude MD文件，里面包含了指令和上下文信息。

But at the highest level, we have this Claude MD file that has, you know, the instructions, the context.

我马上分享一下这个文件。

I'll share that in a second.

这基本上就像是Claudia的工作职责描述。

Basically, kind of like the job description that Claudia has.

然后我们有一组命令，这些命令会与Claudia一起执行。

Then we have a list of commands that basically run with Claudia.

比如，如果我们想在数据收集当天进行质量检查，或者想要每周了解各客户的情况进展，或者正在为新客户搭建系统，又或者在入职新人时。

So if we wanna do a quality check on the day that that that is collected, if we want a weekly update on what's going on across clients, we're trying to set up a new client, if we're onboarding someone new.

然后我们还有一系列任务。

Then we have a list of tasks.

任务功能是最近才推出的，它们对Claudia的高效运作起到了关键作用，因为任务是分阶段进行的。

Tasks came out fairly recently, and they have been instrumental to Claudie being effective because they are tasks happen in phases.

它们管理依赖关系，并允许子代理在工作返回给我们之前进行双重甚至三重的质量核查。

And so they manage dependencies and enable sub agents to basically double check and triple check the quality of the work that's being done before it comes back to us.

然后我们还有一些通用型的代理。

And then we have some general purpose sort of agents.

所以有一些技能文件和通用原则。

So some skill files, general principles.

比如，我们希望内容格式规范。

You know, we want things to be well formatted.

我们希望内容的表达方式能体现我们品牌的风格，这些技能都是在后端启用的。

We want the things to be written in a way that reflects, you know, every in our brand, and those are skills that we've enabled on the back end.

然后这可能是最重要的部分。

And then this is kind of like maybe the most important part.

那就是数据源。

It's the data sources.

对吧？

Right?

我们启用了连接到Gmail、日历、Google Drive以及我们工作相关会议记录的MCP。

So we we enabled MCPs that connect to Gmail, to Calendar, to Google Drive, and into the meeting transcripts for the work that we do.

因此，在最高层次上，这可以说是Claudia或项目经理的架构，这些是一些帮助Claudia有效工作的命令。

So at the high low highest level, this is sort of like the architecture of Claudia or project manager, and these are some of some of the commands to to kind of these are some of the commands that help Claudia work effectively.

所以如果你正在观看这段内容并且需要一个产品经理，这将是一个非常好的模板或模型，可以帮助你思考如何设置一个适合你的项目经理。

So if you're watching this and you need a product manager, this is a pretty good this is gonna be a pretty good sort of template or model for how you can think about setting a project manager that could work for you.

我现在要深入探讨一下我称之为clot md文件的内容，这实际上是我们给Claudia的职位描述。

I'm actually going to now dive into the what I'll call this is the clot m d file, and this is really what I'll call the job description that we've given Claudie.

这是一个Claudia每次我们要求她做事时都会阅读的文件。

So this is a file that Claudia reads every single time we ask her to do something.

我们发现这非常重要，因为如果你是一名项目经理，你总是清楚自己在哪里工作、自己的职责是什么、什么是优秀的工作表现、向谁汇报以及同事是谁。

And we found this to be really important because if you are a project manager, you always know where you work, what your job is, what it means to do a good job, who you report to, and who your colleagues are.

对吧？

Right?

作为一个人，你总是知道这些信息，所以我们希望创建一个文件，从根本上持续为Claudia提供这些信息，让它记住自己与谁合作，以及可以从哪里获取信息来出色地完成工作。

If you're a person, you always know this information, and so we wanted to create a file that, at its baseline, always gave Claudia this information to remember where who it's working with and where it could be drawing information from in order to do its work really well.

所以它知道我们和谁合作。

So so it knows who we work with.

它知道从哪里获取数据，并且每次遇到仪表板时都知道如何处理。

It knows where to draw data from, and then it knows every time it encounters a dashboard.

这是我们组织信息的通用方式，以便它能够高保真地持续填充和维护这些信息。

This is the general way in which we've structured information so that it can continue to populate and maintain that information with high fidelity.

这里有一些ID命名规范，我会快速提一下，因为如果你要创建自己的项目管理者，我们会发现一些数据库管理原则实际上对项目管理非常有帮助，尤其是在关联不同信息片段时。

Here, there's some ID conventions that I'll actually mention really quickly because if you are creating your own project manager, we realize there are some principles of database management that actually help a lot in in project management where you are relating different pieces of information to each other.

对吧？

Right?

这个人参加了这次培训吗？

Did this person attend this training session?

这次培训是否提供了提示、代理或其他内容？

Did this training session deliver prompts or agents or whatever it was?

因此，创建类似数据库管理的ID规范，让Claudia能够追踪谁在何时何地做了什么，这对我们让整个系统良好运行至关重要。

And so creating ID conventions that are effectively like database management that that allow Claudie to connect who did what where were a huge unlock for us to have this entire system work well.

然后，我们给了Claudia一些需要始终牢记的原则。

And then we gave Claudie some principles to always keep in mind.

所以，数据准确性至关重要。

So, you know, data accuracy, totally key.

你必须主动，而不是被动。

You have to be proactive, not reactive.

别等着别人提醒，如果你想做个好员工，就要主动一点。

Don't wait for you know, if you wanna be a good employee, you wanna be proactive.

对吧？

Right?

你不想被人吩咐着去做事情。

You you don't want to be asked to do things.

所以我们给了它这种心态。

So we kind of gave it that mentality.

每一次互动都在建立或削弱信任。

Every interaction builds or erodes trust.

用公式代替手动输入，这些都是很好的最佳实践，如果你是项目经理或经常处理数据，一定要认真考虑。

Formulas over manual entry these are just good, good best practices that if you are a project manager or you deal with data, you really think about.