我们用20个AI代理取代了销售团队——以下是发生的情况 | Jason Lemkin (SaaStr)

所以这一切都彻底不一样了。

So that just totally changes.

上门拜访、清楚自己卖的是什么、懂得怎么搞定采购流程——这些办法依然管用。

The play still work, showing up in person, actually knowing what the hell you're selling, knowing how to get through procurement, all of those are work.

但除了2020年那些异常的短暂窗口期，我们从未见过这么多人同时有购买需求。

But other than these weird windows, artificial windows in 2020, we've never had so many people in market at once.

这是特指AI产品，还是说所有产品都这样？

And this is for AI products specifically or Yeah.

那些能带来巨大投资回报的AI产品。

That that have massive ROI.

对。

Yeah.

提升效率。

Productivity.

我

I

我想在我们公司引入一个氛围代码工具，Lenny。

want I I need to bring a vibe code tool into my company, Lenny.

去吧，去试试Replit、Lovable还有其他工具，然后买一个。

Okay, go out and do the work, compare Replit, Lovable and whoever else and buy one.

明白吗？

Okay?

为什么Harvey和其他人，还有Gora做得这么好？

Like why Harvey and the others and the Gora doing so?

我的意思是，这些工具确实不错，但现在大家都觉得，得用AI来自动化合同和文档的审阅了。

I mean, they're great tools, but everyone's like, we need to automate how we review contracts and documents with AI now.

他们需要一个领头的解决方案，而且会主动去落实。

They want a leader and they're going to do it.

但这种势头迟早会慢下来。

And that will slow down.

不是每个人每年都能挤进市场。

Like not everybody can be in market every year.

这会让企业精疲力尽。

It exhausts an enterprise.

所以这波AI热潮终会过去，某种程度上我们会回归传统做法。

So this is a version of the AI bubble that will end and we will revert in some ways to old school.

但当所有人都涌入市场时，整个运作方式就彻底变了。

But when everybody's in market, it just changes how you run the whole thing.

所以增长最快和最慢的公司，都有各自的理由去使用AI。

So the so the fastest growing ones and the slowest growing ones both have incentives to use AI here just for different reasons.

那销售这个职业呢？

How about the sales profession specifically?

销售开发专员会被客户经理完全取代吗？

Are SDRs gonna be replaced fully AEs?

你如何看待销售职业的未来？

How do you see the future of the sales profession?

那种刚毕业就招来、负责发邮件、回复客户来信、可能当天或第二天再回电的初级销售专员，我们根本不需要了。

The classic SDR junior kid that is hired out of college to send emails and respond to inbound emails and maybe get back to them later that day or the next day, we don't need them.

我们不需要大多数人了。

We're not going to need most of them.

在很多行业里，上门拜访的销售助理并不会被取代，对吧？

SDRs that knock on doors in a lot of industries aren't going to be displaced, right?

明年，靠邮件跟进的销售助理有90%会被AI取代。

The email based cadence SDR will be 90% displaced by AI next year.

不同公司对这些岗位的叫法不一样。

The people have different nomenclature.

我管那些筛选潜在客户的人叫BDR，就是处理我们收到的联系请求的，现在根本用不着他们。

I call BDRs, folks that qualify leads coming in, the contact needs that we see, we have no need for them today.

明年他们就该彻底消失了。

They should be extinct next year.

在AI时代，我何必点个‘联系我们’，然后等一两天，让一个连线性代数是啥都不知道的21岁年轻人回我一句‘你干嘛的？’

There is no reason in the age of AI, I have to hit contact me, wait a day or two for a 21 year old that doesn't know what linear does to say, hey, what do you do?

你愿意为我付多少钱？

How much are you willing to pay me?

我这周晚点可能跟伦尼约个电话。

Maybe I'll set up a call with Lenny later this week.

有AI的话，根本不用做这些。

There is no need to do that with AI.

我们的AI就够了，其中一个智能代理就能自动完成网站上所有访客的资质审核，对方甚至都察觉不到。

Our AI alone, one of our agents, fully qualifies everybody on the website so they don't even know they're being qualified.

它直接帮他们安排好和销售的会面。

It just sets up the meeting with the salesperson.

所以这种邮件型销售代表，还有人工筛选线索的做法，对客户来说其实挺糟糕的，被人筛选的感觉并不好，对吧？

So this SDR, this email based SDR and this human qualifying leads, which is not good for the customer, it doesn't feel good to be qualified, does it?

明年这时候，这些岗位基本就没了。

They will be mostly extinct next year.

我猜你那边的客户经理，就是传统那种做销售的人，现在大部分工具还没跟上。

I'm guessing with your, now the AE, the classic human doing the sales, most of the tools aren't there yet for the most part.

我觉得到明年年底，他们70%的工作还能保住，但之后可能会降到四成到五成。

I think 70% of their jobs will be safe by the end of next year, but I think it will decline to 40 or 50.

我觉得，我们在其他领域看到的情况表明，一个优秀的智能代理完全也能促成交易。

I don't think there's any reason what we're seeing in other categories, a great agent can't close a deal too.

如果价格没什么可谈的，而且代理对产品的了解比人类还透彻，至少对你我这样的人而言——你愿意跟销售员真人聊天吗？

If there's not a lot to negotiate on price and the agent knows the product better than a human, at least for folks like you and me, I mean, do you like to talk to a human in sales?

有时候吧。

Sometimes.

我更喜欢跟一个特别聪明的AI聊天。

I'd rather I'd rather just chat chat with a, yeah, a really smart AI.

没错。

Yeah.

这些现在都在推进中。

So that's all in progress now.

但有个经典又棘手的问题，很多人会问，Lenny，他们会说：‘杰森，我明白你的数据，可问题是……'

But the classic and the tough part, a lot of folks ask this question, Lenny, they say, okay, Jason, I I see that in in your data.

如果连SDR和AE这样的初级销售岗位都没了，那销售这个职业该怎么延续？

How are we gonna build the sales profession if there's no entry level jobs in SDRs and AEs?

这其实是个贯穿整个AI领域的深层问题。

And that's a meta question across all of AI.

我们已经看到AI让中层人才变得更强大了，对吧？

We're already seeing AI concentrate strength in sort of mid tier folks, isn't it?

现在很多人都在削减初级岗位的招聘，除了Shopify这类公司，他们更愿意雇用六七年经验的工程师，而不是去培训新人。

And we're already seeing lots of folks cut back on entry level hires, you know, Shopify and others aside, where they'd rather have the six or seven year old engineer that's a cursor machine rather than train some kid.

如今这么做效率高多了。

It's just more efficient today.

你说是不是？

Right?

我相信你在跟很多人交流时也注意到这一点。

I'm sure you see that across a lot of folks you talk to.

销售领域也会出现同样的趋势。

It's going to happen in sales too.

所以，那些懂得如何管理AI代理、真正了解产品的人都会变得更重要，其他人则会逐渐失去优势。

So the folks that know how to manage an agent, work with an agent, the folks that know their product for real, they're going to become more valuable and the rest are going to become less valuable.

挺有意思的，因为我也投了不少初创公司，你应该也是。

It's interesting because I'm an investor in a bunch of startups, as are you.

我确实看到很多公司在招市场和销售方面的人。

And I'm actually seeing a lot of asks for go to market people, salespeople.

你觉得这只是一阵子的事吗？因为需求太大，他们觉得需要人手帮忙，但以后慢慢就会更多依赖AI了？

Do you think this is kind of a temporary because there's so much demand, they're like, oh, we need people to help, and then this will start to become more AI over time?

还是说他们其实就是在找你说的那种那种资深人士？

Or are they just looking for these really senior people that you're talking about?

话说回来，不管你是管人还是调度AI代理，领导力都是必不可少的。

Well, listen, whether you're managing humans or orchestrating agents, you need leadership.

目前我们还没见过能自主运作的CEO呢。

We've yet to produce an autonomous CEO.

我知道推特上有人讨论AI会取代CEO，但我觉得这更多是比喻说法，而不是字面意思，对吧？

I know folks talk about how on Twitter that will AI will replace the CEO, but I don't know if that's literal as much as figurative, right?

所以，我们依然需要那些高管团队。

So we're still going need the c suite.

我们仍然需要副总裁。

We're still going to need VPs.

我的意思是，管理一百万条线索的工作量已经大到不行了。

I mean, it's become so much work to manage a million leads.

对吧？

Right?

五十万条线索也是一样。

A half million leads.

所以我们还是需要这些领导者，至于问题在于——我看到过你的发言，也看过你发的那条推文。

So we we need these leaders whether the the question is, and I've seen your call out there and I saw your tweet on it.

问题是，那些现在还在用旧方法的人，有多少真正能适应未来？

The question is, how many of the folks that had the current playbooks are the right folks for the future?

我觉得我接触的人里，大概只有两成吧。

I'm thinking maybe 20% of the folks I talk to.

有两成的人还在为人工智能感到恐慌。

20 are still panicking about AI.

如果你想了解怎么做，我就告诉你怎么成为那20%的人。

And I'll tell you how to not, how to be in the 20% if you want to know.

但我认为，像Vercel的珍妮这样的人，能跳过去的寥寥无几。

But I think very few like the Janine from Vercel are going to make the jump.

咱们走着瞧吧。

So we'll see.

肯定会有大型企业这么做，对吧？

There will be huge organizations, right?

比如丹妮丝，她在Slack和Salesforce干了十四年，现在去OpenAI当首席营收官了。

Like Denise just went from Slack and Salesforce after fourteen years to be CRO OpenAI.

她的职位会非常高。

She's working gonna be pretty high level.

对吧？

Right?

所以她

So she

可能还不知道该怎么去实现这些智能体。

may not know how to need to know how to implement the agents.

但你公司想招的大多数人，我建议你确认他们真愿意撸起袖子，干2026年、2027年这活儿。

But most of the folks your companies wanna hire, I would just make sure they really want to roll up their sleeves and do the job of 2026, 2027, right?

在Slack干过，并不等于就具备你初创公司需要的本事。

Just because they worked at Slack does not necessarily mean they have the skills at your startup.

你之前说有一些方法，能让人真正成为那20%。

You said that you had some tips for folks to actually be this 20%.

如果有人听到了，心想‘糟了，我的工作要被取代了’，那该关注什么？

What are some If someone's listening to this oh man, my job is in trouble.

你应该重点做什么？

What should I focus on?

听起来可能很简单。

It's going to sound simple.

但真的管用，而且几乎没人这么做。

It will work and almost nobody is doing this.

选一个工具、一个智能代理，或者一个能帮你解决某个问题的智能系统。

Pick a tool, an agent, an agentic tool to solve one of your problems.

具体选哪个其实没那么重要。

It doesn't almost matter.

就挑那个让你最头疼、最迫切需要解决的，比如客服、销售开发，或者潜在客户筛选都可以。

Just one that's the most painful or the one that's most acute could be support, it could be SDR, it could be inbound qualification.

选一个就行。

Pick one.

选一家领先的供应商。

Pick a leading vendor.

哪个都行，我不在乎。

I don't care which one it is.

明白吗？

Okay?

关于怎么选供应商，我们后面可以聊，但先选一家服务好、你用着顺手的，自己动手试试。

We can talk about how to pick a vendor, but pick a leading vendor that treats you well, that you like, and do it yourself.

训练这个智能体。

Train the agent.

导入数据。

Ingest the data.

不断迭代。

Do the iterations.

搞清楚这玩意儿到底是怎么运作的。

Understand how this damn thing works.

明白吗？

Okay?

今天感到迷茫的人，根本没亲自做过。

The folks that are lost today have never done it.

我们真的变成咨询公司了，Lenny。

We literally we've turned into a consulting shop, Lenny.

这简直有点荒唐。

It's kinda crazy.

我也不太确定自己怎么想这件事。

I I don't know what I think about it.

但我们刚刚确实做了一个项目。

But we literally just did a job.

我们的首席AI官阿米莉亚和米娅负责操作的。

Amelia, our chief AI officer, and Mia, she drove it.

我们跟一家市值超过一千亿美元的B2B公司开了会，你本以为他们是AI领域的领军者。

We did a call with a public b to b company worth well over 10,000,000,000 that you would think is an AI leader.

明白吗？

Okay?

我们跟他们的团队开了会，他们说：我们完全懵了。

And we did a call with their team and they're like, we're struck.

他们想搞清楚这个AI销售助手是怎么回事。

We wanna figure out this AI SDR stuff.

这挺好。

Like, great.

我们还以为只要买个工具，扔给销售开发人员，他们自己就能搞明白。

We think we're just going to buy this tool and just give it to our SDRs to figure it out on their own.

第一，根本不可能。

Okay, one, no chance.

第二，我们问他们，你们自己实际操作过多少？

Two, we asked them, how much of this have you done yourself?

你们自己有没有亲手做过？

Like, have you done it yourself?

结果在这场二十个人的电话会议里，全场鸦雀无声。

And it was just crickets on this call of 20 people.

没人真正亲自做过。

No one had done it their self.

他们以为只要把一个没受过训练的工具丢给一群二十岁出头的销售开发人员，它就能自动成交。

So they thought they could take an untrained agent with no training and just magically give it to a bunch of young 20 year old SDRs and this magically would sell on its own.

事情哪有这么简单，对吧？

It doesn't work that way, right?

这些智能代理的工作方式其实有很多术语，听起来挺吓人的，比如数据导入、流程编排、模型训练之类的。

So, the way all these agents work is you there's a lot of jargon which is intimidating, ingestion, orchestration, training.

其实没那么复杂，各位。

It's not that hard guys.

只是跟以前不一样罢了。

It's just different.

说到底，还是我们过去十多年一直在做的那些B2B销售老一套。

It's the same B2B stuff we've been doing for over a decade.

你先去一个网站，明白吗？

You go to a website, okay?

把你们公司官网的链接发给它。

You give it a URL of your website.

再把你们维基页面的链接也给它。

You give it a URL of what your wiki is.

然后把培训资料的链接也一并提供。

You give it a URL of your training docs.

也许你上传一份说明书。

Maybe you upload your prospectus.

你上传几份文件。

You upload a few documents.

它会把数据导入进来，所谓导入就是上传的意思。

It ingests the data and ingesting means it uploads.

它会处理这些数据，还会做些别的事，但你其实不用关心，什么向量化、吹牛之类的，根本不重要。

It means it processes the data and it does some other stuff you don't really need to know, some bragging, some vectoring, it really doesn't matter.

你上传一些东西，它大概能懂一点，但还不太行。

You upload some stuff and it kind of knows it and isn't great at it.

然后它会试着把这些内容转化成问题。

And then it will turn it into, ideally it will turn it into questions.

你来回答这些问题，答得越多，它就越能学会。

And you answer these questions and they will be get, the more you answer and train it.

训练其实就是不断回答问题，让它越来越精准。

Training is just answering questions and getting better and better.

首先你上传一堆资料，然后花好几个小时训练它，通常还得靠供应商帮忙，有个叫‘前线部署工程师’的角色，听着挺吓人。

So first you upload a bunch of your stuff, then you spend hours training it, often with the help of a vendor, someone called the forward deploy engineer, which a scary term.

其实就是有人来帮你完成这一步。

It means someone is going to help you do this.

你上传资料，尽量弄对，之后还得确认它真的没问题。

You upload your stuff, you try to get it right, and then you basically have to make sure it's right.

质量检查，测试。

QA, testing.

每天当这个AI销售助手发邮件、练习邮件时，总会说些蠢话。

And every day when that AI SDR sends out emails and do practice emails, they will say some dumb things.

可能是它瞎编的。

Maybe it's hallucinations.

技术术语叫什么其实不重要，你发现错了就纠正它。

It really doesn't matter what the technical term is and you correct it.

如果你坚持三十天，每天花一两个小时纠正这些错误，到第三十天，它就会变得相当靠谱了。

And each, and if you do this for thirty days, then every day you spend an hour or two correcting those mistakes, by the thirtieth day, it's going to be pretty good.

只要是做过B2B或者SaaS的人，谁都能做到这一点。

And this is anyone can do this that has been in B2B or SaaS.

我刚才说的这些，任何人都能做。

Anyone can do what I just described.

这跟我们以前做过的事其实没多大区别。

It is not that different than other things we've done.

只是步骤安排得不一样而已。

It's just sequenced differently.

但就是没人这么做。

But nobody does this.

大家都慌了。

Everyone's panicked.

如果你真想试试，随便选个工具就行。

And if you if you can go do this, pick any tool.

选Agentforce，选Qualified，选Artisan，或者任何你感兴趣的都行。

Pick Agentforce, pick Qualified, pick Artisan, pick whatever you want.

如果你能动手把这个做出来并上线运行，那你就是抢手人才。

If you can go do this and get it live into production, you're hyper employable.

所有你提到的那些需要搞增长的人的公司，都会抢着聘你。

All the companies you talked about that need a GTM person, they will hire you.

你完全可以成为他们的首席智能增长官，因为只要愿意，几乎人人都能做到。

You could be magically be their chief agentic GTM officer because almost anyone can do this if they want to.

这大概只需要你花一个月时间，加上五十到六十个小时，再加点时间筛选供应商。

It's just going to take a month of your time and it might take you fifty or sixty hours plus qualifying the vendor.

对吧？

Right?

以前呢，比如我们刚做播客那会儿，你直接外包给一家公司，然后就不管了。

And in the old days, like when we first did our podcast, you'd hire an agency and disappear.

过去大家都是这么干的。

That's how you do this stuff.

不是说那些代理公司不会做，而是他们根本不知道该怎么操作。

Don't it don't work that the agencies don't know how to do this.

你得亲自上手做。

You've got to do it yourself.

但只要你做了，真的会特别厉害。

But if you do, man, you will rock.

你肯定会特别出色。

You will just rock.

而且你会学到很多。

And you will learn.

对吧？

Right?

你会学到很多。

You will learn.

还会明白其中的界限。

And you will learn the limits.

知道这个工具能做什么，不能做什么。

And you will learn the agent can what it can do and where it can't do.

然后你就会学会怎么处理下一个任务，对吧？

And then you will learn how to do the next one, right?

我们现在已经用上了Agentforce，就是Salesforce的那个系统，Mark经常提到它，但像我们这种规模的公司，可能没几家在用。

So like we're pretty far on Agentforce, which is Salesforce's one, which Mark talks a lot about, but we're probably one of the only organizations of our size on it.

我跟你说个实用的小技巧，挺有意思的。

I will tell you a cheat code, which is pretty interesting.

我们之前有三个这样的智能代理在负责销售工作。

So, we've got we had three of these agents working for sales.

经过训练、摸索，花了几个月时间，最后我们把所有操作都浓缩成一个提示词了。

After training it and learning it and spending months time to learn it, we got it down to one prompt.

‘提示词’这个词听起来可能有点吓人。

And prompt is another almost intimidating word.

其实就是一段文字，告诉你这个工具该做什么。

A string of text that describes what you want this thing to do.

明白吗？

Okay?

我们把这个提示词交给Agentforce，一天之内它就已经表现得相当不错了。

We took that prompt and gave it to Agentforce, within a day it was pretty good.

所以如果你能做出一个这样的东西，那难度会非常大。

So you will, if you can do one of these, it'll be really hard.

简直会让人崩溃。

It'll be brutal.

但第二个就会简单多了。

Then the second one will be easier.

如果你能自己搞定，那你就是AI领域的顶尖高手了。

And then you're going to be like the master of the universe in AI if you can do it yourself.

但如果你等着团队里的人来做，或者指望外包公司，那我敢说你迟早会丢掉工作。

But if you're waiting for people on your team to do it, if you're waiting for an agency to do it, I think you're gonna be out of a job.

对吧？

Right?

所以现在大家都把我们当成专家来找。

So this is like everyone comes to us as an experts.

我们之所以算专家，是因为我们已经做了二十遍了。

We're not we're only experts because we did it 20 times.

我觉得现在最好先带你快速浏览一下我们开发的那些智能体，看看它们各自做什么，哪些最有影响力。

I think what might be helpful here actually is to do a a tour, a a quick tour of the agents that you've built and what they do, which ones have been most impactful.

在介绍的过程中，也说说你用的是哪些产品。

And then as you do that, what products you use?

哪些工具支撑了这些智能体？你喜欢的和不喜欢的分别是什么？

What powers these agents that you like and maybe don't like?

如果有人访问 saster.aiagents，就能看到我们做的所有东西。

If anyone goes to saster.aiagents, we'll you'll see everything we built.

全都列得清清楚楚。

It's all bulleted out.

你可以直接照着我们的做，我来一步步带你过一遍。

You can copy us and I'll walk you through it.

不过有两点得提前说明一下。

But two two caveats or things like that.

我在 Replit 上做了很多东西。

I built a lot of stuff in Replit.

我们可以随便聊聊，就当娱乐了。

We can talk about it for fun.

我算是排名前1%的用户了。

I'm like a top 1% user.

我特别喜欢这个平台。

I love it.

我们做的所有市场推广相关的东西，都不是自己开发的。

None of the GTM stuff we built ourselves.

别自己动手做这些了。

Don't build it yourself.

你又不是 Vercel。

You're not Vercel.

你也没有全职的顶尖工程师愿意花时间去开发这种东西。

You don't have a full time wicked awesome engineer that wants to build this.

这些东西真能自己搭出来吗？

Could all this stuff be built yourself?

这跟自己做Notion是一个道理。

It's the same idea of building your own notion.

你能做，但别做。

You could do it, but don't do it.

这些产品价格不便宜。

It's these products are expensive.

也没那么贵。

They're not so expensive.

值是值，但维护起来麻烦。

It's worth and then maintaining.

创新的速度太快了。

The pace of innovation is so fast.

就算你能请人内部开发，要是不留意，过几个月就过时了。

Even if you can hire someone to build it internally, it will become obsolete if you're not careful in a couple months.

所以我们已经做了不少东西了。

So, we've built a lot of stuff.

我曾经写过一个计算器，用来做创业相关的计算，九十分钟内用了八十万次。

We could talk about I built a calculator to do startup calculations to use 800,000 times in ninety days.

用于评估。

For evaluation.

对。

Yeah.

用于评估。

For evaluations.

我还开发过一个商业计划书评审工具，已经审过将近三千份商业计划书。

I built a pitch deck reviewer that's reviewed almost 3,000 pitch decks.

做了不少有趣的东西，但所有市场推广相关的工具，我们都没自己做。

Lots of fun stuff, but none of the GTM stuff we built ourselves.

一样都没自己做。

None of it.

所以先提醒一下，别自己动手做。

So just a caveat, don't build it yourself.

除非你是Vercel，或者有特别的理由。

Unless you're Vercel, unless you have a reason.

那期播客真不错。

That was a great pod.

特别棒。

It was wonderful.

别这么做。

Don't do it.

别干这种事。

Don't do that.

说白了，除非你有特别出色的市场工程师。

Basically, if you unless you have a a awesome go to market engineers.

而且他们真的很想做。

For and they really wanna do it.

他们早就迫不及待想动手了。

They really are chomping at the bit to do it.

别这么做。

Don't do it.

我一开始是这么做的，虽然别人不会这么开始，但确实有些经验可以分享。

So I started and this is not where other folks would start, but there's some learnings.

我一开始用了一个叫 Delphi 的应用，它可以创建数字分身，你以前给 Lenny bot 用过这个工具。

I started there's an app called Delphi, which makes digital clones, and you used it for the Lenny bot a long time ago.

我看到你用过。

I saw you do.

那是什么？

What's that?

是的。

Yeah.

Lennybot.com。

Lennybot.com.

是的。

Yeah.

去看看。

Check

试试。

it out.

我很久以前就看过。

I saw it a long time ago.

挺有意思的，但我就是没搞明白。

It was interesting, but I I I didn't it didn't click.

后来HubSpot的创始人兼董事长布莱恩·哈利根也做了一个。

And then Brian Halligan, who's the founder and chairman of HubSpot, did one too.

他们有红杉资本支持，当时布莱恩在红杉工作，所以很早就帮了他们。

They're Sequoia backed, he was working at Sequoia, so he helped them early on.

然后我突然顿悟了，AI就是这样，当我看到这两者结合的时候。

And then I kinda had a magic moment, and this is the way it works in AI, when I saw the combination of the two.

所以你的这个工具真的很棒，要是还没试过Moneybot，真该去试试。

So yours was real like, people should love Moneybot because it's got it's if folks haven't tried it, tried.

它会自动吸收数据。

It is ingested.

对有些人来说，这个词听起来挺吓人的。

I know a scary term for some.

你所有做过的访谈，它都会全部吸收进去。

It is ingested every single interview you've ever done.

对吧？

Right?

你写过的每一篇内容，它也都收进去了。

Every word of content you've written.

它能把这些全都整合起来。

So and it can combine them all together.

它能把Vercel的故事、你和Dylan在Figma的合作经历，全都融会贯通，生成的内容相当出色。

It can combine the Vercel story and the and and and and and what what you did with with Dylan at Figma, and it can synthesize the knowledge, and it's pretty good.

挺不错的。

It's pretty good.

不过我更喜欢布莱恩那个，因为那就是布莱恩本人。

What I liked about Brian's better than yours, though, was that it was Brian.

Lenny Bot 有点像 Lenny，但又像是你所有嘉宾的合体。

And Lenny Bot is kinda Lenny, but it's also kind of all your guests.

对吧？

Right?

我觉得这才是它最厉害的地方。

I think that's the superpower of it.

对吧？

Right?

我也是这么想的。

That's the way I think about it.

这不只是我自己的想法。

It's not just my intelligence.

这档播客里每个嘉宾教给我的东西，都融进了这里面。

It's the the lessons of every single person I've had on this podcast.

所以这真的很棒。

So it's great.

但我突然想，也许我可以做一档介于两者之间的节目。

But I thought, hey, maybe I could finally do one that's that's in between the two.

我当过创始人，也写过上万篇内容。

Like, I I've been a founder and I and I've written 10,000 pieces of content.

这点有点像Brian，但我比他更丰富。

So that's a little bit like Brian, but I have more than Brian.

我在效率上赶不上Lenny，但我有更多元的声音。

And I'm not Lenny in terms of productivity, but I've got a lot of voices.

所以我决定试试看。

So I'm like, I'll try it.

我用了Delphi。

I used Delphi.

我一用就崩了，因为数据量太大，根本处理不过来。

I instantly broke it because I had too much data to ingest.

花了大概一周时间才总算跑起来。

It took about a week to get going.

后来它真能用了，用户反馈也跟你说的一样。

And it worked and people and it's just like you.

有些人每天花好几个小时在别的浏览器里查JSON，但他们不会像用Lenny那样做事；他们会问销售上的难题，该怎样带团队，还会上传领英资料，问该不该招人。

People some people spend hours a day on digital JSON in another browser and they don't do what they do with Lenny, but they'll they'll ask about their sales woes and what to do with their sales team, and they'll and they'll upload LinkedIns and ask if they should hire people.

后来发生了一件挺有意思的事——因为我们办活动，大家开始用它来问跟活动相关的问题。

And then a curious thing happened, which is that because we do these events, people started to use it for questions for the events.

是的。

Hey.

我怎么申请退款？

How do I get a refund?

能给个折扣吗？

Hey, can I get a discount?

嘿，圣马特奥县博览会场地在哪儿？

Hey, where is the San Mateo County Fairgrounds?

所以杰森·莱斯真的在旧金山湾区。

So Jason Lies is really in the San Francisco Bay Area.

诶，刚才谁在说话？

Like, who's speaking?

而且问题多得没完没了。

Or and and like, there's endless questions.

对吧？

Right?

以前我们用的是Pre-Fin Intercom，忙得根本顾不上，经常两周后才回复。

And we used to use pre Fin Intercom, and we're so busy we would answer like two weeks later.

说真的，那简直是史上最差的客服。

Like, mean, was worst support ever.

后来那个客服代理居然自己开始独立处理问题了。

And the the agent just started doing support on its own.

然后它自己就开始卖广告赞助了。

And then it did this thing where it sold sponsorship on its own.

对吧？

Right?

所以如果你还没开始，不妨先从客服入手。

So so start you can start with so one place to start if you haven't started is in support.

明白吗？

Okay?

你不一定非得买Sierra，也不一定非得买Decacon或者Finn，但有个不错的起点就是优化你的客服系统。

And you don't have to buy Sierra and you don't have to buy Decacon necessarily and you don't have to buy Finn, but one potential place to start is is your support.

你能提供全天候的优质客服吗？

Like, can you do great twenty four seven support?

你做得到吗？

Can you?

大多数应用都做不到。

Most most apps can't.

事实上，一些最差劲的案例恰恰来自AI领域的领军者。

In fact, some of the worst offenders are AI leaders.

他们的网站上根本没有任何客服支持。

They they have no support at all on their website.

所以，这是一个很好的切入点。

So that's one place to start.

接下来，我们开始考虑另一个方向。

And then the next place we started okay.

对我们来说，下一个尝试的方向是主动出击。

So for us, login for the next place we started is, hey, we wanna try outbound.

明白吗？

Okay?

我们没有像你们那样上百万的用户名单，但也有大概四十万。

Because we don't have 1,200,000 names like you have, but we have like 400,000.

明白吗？

Okay?

我们掌握着他们的数据。

And we have data on them.

所以我们想邀请他们再回来参加我们的SaaStr活动。

So we wanted to say, hey, come come back to our SaaStr event.

当时我们也不确定该用什么工具，我跟你说件事。

So we didn't know what to use, and I'll tell you something.

我们选了一家YC的公司，叫Artisan。

So we picked a YC company called Artisan.

他们今年从零起步，做到了一千万的规模。

They've gone from nothing to 10,000,000 this year.

我们选了他们，但这里有个关键原因。

We picked them, but this is important why.

他们曾经是SaaStr的赞助商，我们原本不知道，但他们主动提供了最大的支持。

They were a sponsor at SaaStr, we didn't know, and they offered to help us the most.

这才是最重要的洞察。

This is the critical insight.

我们当时并不确定Artisan是不是真的如我后来所想的那样。

We didn't know if Artisan was the I have opinions now.

我们还没用上它，但另一个供应商的首席营收官跟我们争了起来。

We hadn't deployed it, but another vendor, the CRO argued with us.

他说：‘你们得先付我十万美金，我才会帮你们。’

He said, I need a 100 k upfront before I help you.

明白吧？

Okay?

还有一个说他们害怕跟SaaStr扯上关系。

Another one said they were scared of SaaStr.

怕万一失败了，会带来负面舆论。

They didn't want bad PR if it failed.

这说得通。

Fair.

哦，但确实如此。

Oh, but yeah.

就是那家特定的公司

Like, that certain company

我们不想成为你的第一个客户

We don't wanna be your

第一个。

first.

明白吗？

Okay?

Artisan说，我们来做。

And Artisan said, we'll do it.

而我们当时什么都没有。

And we had nothing.

关于Agentics的东西，有个有意思的地方。

And here's the interesting thing about Agentics stuff.

就是售后服务。

It's support.

如果你什么都没有，也没必要非得改变世界。

If you have nothing, like it doesn't have to like change the world.

你要是真的一无所有，但开始做一件投资回报率高的事，那肯定能有回报，对吧？

If you're literally doing nothing and you start to do something that's high ROI, like you're gonna get return, right?

所以我们先做了这个，训练了一下，效果很好。

So, we did that one, we trained it, it's great.

我们发了大约六万封邮件，转化率相当不错。

We did about 60,000 emails, saw pretty high rates.

然后我们就想，不如试试主动跟进的方式。

Then we said, well we'll try inbound.

我们不希望出现销售一走，客户就得等两周才能有人联系的情况。

Don't want to have this depressing experience where a salesperson quits and it's two weeks later until they talk to it.

于是我们用了这家叫Qualified的供应商，创始人是前Salesforce的首席营销官，现在他们做很多业务，但主要还是专注在Qualified这块。

So we used this vendor called Qualified, which was founded by the ex CMO of Salesforce that does a lot now, but mostly focuses on Qualified stuff.

这个方案一上线就见效了。

That immediately worked.

比如，有个客户在周六晚上11点还想赞助，结果一联系就成了，效果特别好。

Like, we had someone at 11PM on Saturday night that wanted to sponsor, and they sponsored and it worked and it and it worked great.

但话说回来，是他们帮了我们。

But again, they helped us.

这个客服人员正在给潜在客户发邮件，说服他们进行赞助。

And this is an agent that is emailing with prospects, selling them on a sponsorship.

你只要去 saastrannual.com 看一眼，任何人都该买这种产品。

Literally, if you go to saastrannual.com, and anyone should buy a product like this.

不一定要是经过筛选的，但会弹出一个 Intercom 的聊天窗口，你知道吧？

It doesn't have to be qualified, but but and there'll be the bubble, the intercom like bubble Okay.

调整一下。

Got tune it.

用来筛选来自网站的潜在客户。

To to qualifying inbound prospects.

有些人会主动说：嘿。

Folks that say, hey.

我想赞助伦尼的播客。

I want to sponsor Lenny's podcast.

不好意思。

Sorry.

2028年之前的赞助位都满了。

We're sold out through 2028.

不过如果你愿意排队等候，可以在这里登记。

But if you wanna be on the wait list, sign up here.

明白吗？

Okay?

或者对我们来说更好的是，能筛掉那些不合适的人。

Or or even better for us, it would qualify folks that out that weren't a good fit.

对吧？

Right?

这样能省下大量时间，而且24小时不间断运作，直接帮你安排好会议。

It would save so much time and it would do it twenty four hours, then it would just set up the meeting.

所以，第二个方案之所以特别棒，是因为根本没人愿意做这件事。

So, the reason that was a great second one was because no one was willing to do that.

没有一个人愿意打电话去跟这些人沟通。

No human was willing to pick up the phone and talk to these people.

这简直是唾手可得的机会。

So, it was such low hanging fruit.

但前两个方案的关键在于，它们提供的帮助最多——如果你要选一个代理人，这一点最重要。

But the key to the first two, and if you're gonna pick an agent, is they offered to help the most.

说到底，Lenny，这些系统全都在跑 Cloud 4。

At the end of the day, Lenny, these are all running on Cloud four.

它们本质上都是把一堆 API 拼在一起用。

They're all basically using a bunch of APIs mashed together.

这种做法在软件行业一点都不新鲜，对吧？

That's not new to software, right?

在后台把一堆 API 组合起来，早就不是什么新鲜事了。

Mashing a bunch of APIs under the hood.

但说实话，我不希望惹恼任何人。

But deep down, I don't want to get any but trigger anybody.

很多AI销售领域的领头羊，其实彼此之间非常相似。

Many of these leader, the leaders in AI GTM, they're more similar than different.

它们在底层机制上都差不多。

They're more similar than different under the hood.

这并不意味着功能上完全持平。

It doesn't mean the features are our parity.

因为需要培训，而且得花一个月时间，再好的软件，如果没人帮你培训，也不该有99%的人去买。

So, because you have to train them, because it takes a month, the world's best software with no help training you is not 199% of people should buy.

过去，我们会先筛选出最好的软件。

So today, in the old days, we would qualify the best software.

我们会做个表格，逐一对比功能，仔细评估。

We would make a matrix and we would do their thing and we would compare features and do it.

你得再加一列：看看你的现场工程师、解决方案架构师或者客户成功经理，问问他们谁更愿意帮你。

You got to do another column, which is your forward deployed engineer or solution architect or your SE, and talk to them and say, who is going to help me?

在签支票之前，先给伦尼打个电话，确认他是不是真会帮你安排会面。

And before you write a check, get on the phone with Lenny and see if Lenny is really going to do the appointment.

如果伦尼很靠谱，但另一家供应商产品更好却不愿提供支持，那就别选他们。

And if Lenny rocks and the other vendor is better and won't help you, don't do it.

我们早期能这么成功，是因为前两家合作的都是初创公司，对吧？

And that's why we've had so much success is the first two we did, yeah they were startups, right?

所以他们更拼，更卖力。

So they worked harder.

但那些没选中的供应商，根本没跟我们配合过。

But our disqualified just did the work with us.

我们不是不懂，但那确实得花功夫。

And we're not stupid, but it was work.

我们当时真的需要帮助，对吧？

We needed help, right?

所以我学到的是，你得看清楚这个合作伙伴——也就是现场工程师和供应商，很多其实根本不会接你的单，除非他们觉得能帮上你。

And so that's what I learned is you have this partner, the FD and the vendor, a lot of them actually might not take your business if they don't think they can help you.

最厉害的那些，现在反而会拒掉不少生意，挺有意思的，对吧？

The best ones turn away a lot of business today, which is interesting, Right?

从这件事里学到的一点是，很多人一用这类工具，比如Lenny，就会说你数据太多了。

Interesting learning from this for folks is a lot of folks say, and they would say to you if you use these, Lenny, they say you have too much data.

SaaStr跟我们不一样。

SaaStr is not like us.

我们是个初创公司。

We're a startup.

我们规模很小。

We're tiny.

你数据库里有四十万人。

You have 400,000 people in your database.

Lenny那边有一百二十万人。

Lenny has 1,200,000.

这可不是说我只有三百个客户那么简单。

It's not like that's I got I only have 300 customers.

明白了。

Okay.

或者200个客户。

Or 200 customers.

我学到的是，这种想法是错的。

What I've learned is that's wrong.

如果你有300个客户，那有多少人曾经访问过你的网站？

If you have 300 customers, how many folks have come to your website ever?

三万？

30,000?

你有多少潜在客户？

How many leads do you have?

你的数据库里有多少人？

How many folks in your database?

你之前尝试联系过多少人？

How many folks have you tried to reach out before?

这已经超过人力能处理的范围了。

More than a human's doing.

然后突然间，他们会想到：你在HubSpot里到底有多少人？

And then all of a sudden they have a moment like, how many folks do have in your HubSpot?

对吧？

Right?

你在CRM里有多少人？

How many folks do have in your CRM?

他们去查了一下，是三万一千人。

They look it up 31,000.

明白了。

Okay.

你有多少人在跟这些人沟通？

How many folks do you have talking to them?

一个都没有。

Zero.

你不需要像我们这样庞大的数据量，也能让这些智能代理正常运行。

You don't need the scale of numbers that you and I have to make these agents work.

你只需要一点点规模，再加一点流量就够了，没你想象的那么多。

You need a little bit of scale and you need a little bit of traffic, but not as much as you think.

我们积累的这些经验，其实很多人根本不想花功夫去落实。

So all the learnings we have, a lot of folks that honestly, they don't wanna do the work.

他们总说：SaaStr 那么大，数据那么多。

They're like, well, SaaStr has a lot of scale.

人家都做了好多年了。

They have a lot of years.

这话不对。

That's not true.

其实训练方面也是同样的道理。

And it turns out to also be like true with the training.

你肯定也注意到 Lenny 机器人的情况了。

I'm sure you've seen it with Lenny bot.

我本来以为有十二年的内容积累才是关键。

Like, I thought having twelve years of content made the difference.

其实不是。

Nah.

关键是几个月的高质量内容，再加上一个长尾效应，但你不需要像想象中那么多的训练数据，只要够精就够了。

It's having like a couple months of really good content and a long tail beyond that, but you don't need as deep training and as much as you think, you just need a bit to be really good.

所以，只要有点本事，哪怕年收入几百万以上，都能用上这些产品，对吧？

So anyone that has any skill whatsoever, even a couple million of revenue and up can benefit from these products, right?

我们先做了通用机器人，帮我们走到某个阶段，然后做了外呼用的STR，接着是内推，再后来很早就和Salesforce一起做了Agentforce。

So we did general bot, got us to a certain place, then we did STR for outbound, then we did inbound, and then we did Agentforce really early with Salesforce.

一开始我们还真不知道怎么用Agentforce。

And we didn't know what to do with Agentforce at first.

对吧？

Right?

但我们决定去重新激活那些销售觉得不值得花时间的客户。

But we decided we would reactivate the folks that sales decide was not worth their time.

那些联系过销售的人，每个初创公司都这样。

Folks that reached out to sales, and this is true at every startup.

我们刚才还提到一些AI领域的领军人物，有人会说：你知道吗？

We even just talked about some of the AI leaders where a human says, you know what?

我觉得这单提成太少了。

I don't think this is enough commission.

我最近挺忙的。

I'm kind of busy.

我手头正有个四百万美元的Meta大单在谈。

I got a $4,000,000 deal with Meta going.

我们就专门拿这些客户来用Agentforce。

We just took Agentforce just on those.

明白吗？

Okay?

我们用的是非常相似的提示语来训练它。

And we trained it on very similar prompt.

它的回复率达到了百分之七十。

It had seventy percent response rate.

这些人可是急着想和我们互动。

Those are people that were dying to interact with us.

百分之七十已经非常好了。

70% is so good.

这种事情人类根本不愿意做，觉得不值得花时间。

And this is something humans were not willing to do, wasn't worth their time.

我知道这听起来有点苛刻，可能得罪一些销售同事，但现实是，如果你处在线索很多的环境里——我觉得大公司也有线索多和线索少的区别，很多时候根本不够用。

And I know this sounds critical and maybe I'm going to trigger some sales folks, but the reality is you know, if if you're in a lead rich environment, okay, and I think that there's lead rich and lead poor environments for for even big companies besides, like there's not enough.

但你迟早会进入线索过剩的状态。

But you eventually, you become lead rich.

明白吗？

Okay?

销售代表根本不会去跟进大部分线索。

Reps just don't follow-up with a lot of them.

这不过是人之常情。

It's just human nature.

就连你也是这样。

It's even you.

我打赌想赞助这份通讯的人，比你能接受的多得多。

I bet more folks wanna sponsor the newsletter than you can let in.

对吧？

Right?

你真的会挨个打电话给所有人吗？

Do you do you pick up the phone with all of them?

我会回复每一个人，然后告诉他们我们已经满员了。

I reply to all of them, and then we just tell them we're we're full.

但确实如此。

But Yeah.

但你明白我的意思吧。

But you see the point.

对吧？

Right?

就连你的规模，你也明白这道理。

Even your scale, you see the point.

对吧？

Right?

是的。

Yeah.

对。

Yeah.

这变得有挑战了。

It gets challenging.

再想象一下，突然间你有了六个月的库存。

And let's imagine all of a sudden you had six months of inventory available.

我打赌，只要你派个助手自动给那些人回邮件，名额马上就满。

I bet if you spooled up an agent and emailed all those folks back automatically, you'd fill up you'd fill up the docket.

对吧？

Right?

是的。

Mhmm.

其实任何人都能用这类工具，你以为只有大公司才用得上？只要不是小到能亲自回复每一个访客、每一个点击你网站的人，你都能从人工智能中受益。

So anyone can do these sorts of you think you can't unless you're so small that you have sufficient humans to talk with every potential lead, every person that touches your website, every person that clicks with anything, you can benefit from AI.

这大概就是我们的经历。

So that was kind of our journey.

之后我们还做了不少细分领域的尝试。

And then we've done a lot of other niche stuff.

最后我跟你说说我们现在的情况吧，可能学到这儿已经差不多了——我们已经到了几乎没法再往前一步的地步。

I'll tell you at the end where we are today, this is maybe this is almost too much learning is we're at the point where maybe we can't do one more.

因为当初我们刚开始用Delphi的时候，我亲自上手，每天花将近一个小时来训练它。

Because right now, when we did Delphi in the beginning, when I copied you with Delphi, even me, I spent almost an hour a day training it in the beginning.

毕竟一开始用它做客服时，它老是告诉用户错误的日期。

Because when we started to use it for support, it had an initial it started telling people the wrong dates.

明白吧？

Okay?

我们可以聊聊原因。

And we could talk about why.

所以我得去修正它，结果它还是出了一些错。

So I had to fix it and it made some mistakes.

所以当用户开始使用时，我每天早上都得花一小时启动Delphi，检查问题，然后一一回复。

And so when people started to use it, I had to spend an hour each morning firing up Delphi, reviewing the issues, and answering them.

现在我不用再做这些了。

I don't have to do it anymore.

它已经训练得很好了。

It's well trained.

现在有太多代理在运行，邮件量太大了，Amelia每周得花十到十五个小时来审核它们的输出。

We have so many agents going and so many emails that Amelia has to spend, you know, ten to fifteen hours a week reviewing the outputs.

这特别累人，因为代理们通宵工作，周末也干，连圣诞节都不休息。

And it's exhausting because agents work all night and they work weekends and they work on Christmas.

这是个大问题，对吧？

It's a big issue, right?

当 orchestrator 或首席 AI 负责人可不适合偷懒的人，毕竟这些代理从不休息，对吧？

This is not, being the orchestrator or the chief AI person is not a good job for lazy people because the agents never sleep, right?

现在要管理这二十个，花的时间实在太多了。

So it is so much time now to manage these 20.

这挺有意思的。

This is interesting.

我们还不知道什么时候能上第二十一个。

We can't I don't know when we're going to do the twenty first.

可能我们已经满员了。

We may be full.

对初创公司来说，这正是他们得加倍努力的原因。

And for folks that are startups, this is a reason to go harder.

因为今年人人都在抢市场。

Because everyone was in market this year.

明白吧？

Okay?

每个人都会遇到这种情况，而且还会继续发生，但业务流程的调整始终是商业软件的一大难题。

Everyone and it's going to keep happening, but business process change remains an issue for business software.

说到底，很多创始人搞错了，99%的销售根本不在乎业务流程怎么改。

Business process change, at the end of the day and so many founders get this wrong and 99% of sales folks, they don't care about business process change.

销售只关心怎么拿到自己的提成。

Sales folks, they just want to get their commission.

只要价格合理，客户花多少钱买软件其实没那么重要。

Doesn't really matter what you pay for an app for a customer as long as it's fair.

真正麻烦的是，要改变整个业务运作方式，得做多少工作。

It's all the work to do to change the way you do your business.

对吧？

Right?

所以我们现在都已经忙不过来了。

So we're even we're at the point where we're overloaded.

对吧？

Right?

所以你要明白，不管你是个初创公司，还是像Salesforce或HubSpot这样的大厂，未来十二个月内得赶紧把单子签下来。

So just be aware if you're if you're a startup or even Salesforce or HubSpot, maybe maybe close those deals in the next twelve months.

因为那个窗口很快就要关上了——到时候人们会说：听好了，我上个月刚换了五个工具，实在受不了了。

Because the window may close where people say, listen, that's the coolest agent I've ever seen.

我已经被前五个搞累了。

I'm exhausted from the last five.

去年我就得上线五个新系统。

I had to do five last year.

我真的没法再往公司里引进第六个应用了。

I just can't literally cannot bring one more app into my enterprise.

所以这会成为一股逆风，而今天大家还以为所有事情都在顺风前进。

And so that's gonna be a headwind that today everything seems like it has tailwinds.

对吧？

Right?