BuildSpeak每日 builder 文摘
今日归档生词本关于
🎙 播客AI & I by Every· 2026 年 7 月 22 日· 8,261 词 · 约 41 分钟

How Every's Team Used AI to Ship Its Biggest Launch Ever

SPACE 播放 / 暂停·←→ 上一句 / 下一句
Speaker 100:00 - 00:17
We generated the largest subscription revenue increase in Everest's history last week. We did it by launching this thing called the Builder Pack. Basically, we partnered with a bunch of the AI tools we use every day, and subscribers to the Builder Pack get discounts and credits to those tools. So thousand dollars in Codex credits, twelve months of free cursor access,
Speaker 100:00 - 00:17
上周,我们创造了 Everest 历史上最大的一次订阅收入增长。我们是通过推出一个叫 Builder Pack 的东西做到的。基本上,我们和一大批我们每天都在用的 AI 工具达成了合作,而订阅 Builder Pack 的用户可以拿到这些工具的折扣和 credits(额度)。比如价值一千美元的 Codex credits、十二个月免费的 cursor 使用权限,
Speaker 200:17 - 00:17
all that kind of stuff.
Speaker 200:17 - 00:17
] 诸如此类的这些福利。
Speaker 100:17 - 00:35
The Builder Pack is awesome, but I think the actual story behind it, like how we did it, why we did it, how it generated so much revenue, is really interesting. For this episode, I had the team who worked on the Builder Pack. Everything from conception all the way to the full execution at launch come to the show to share how we did it and what we learned. I hope you like it.
Speaker 100:17 - 00:35
Builder Pack 很棒,但我觉得它背后真正的故事——比如我们是怎么做出来的、为什么要这么做、它为什么能带来这么多收入——其实特别有意思。这一期里,我请来了参与 Builder Pack 的团队成员。从最初构想到上线时的完整执行,他们都会来节目里分享我们是怎么做的,以及我们学到了什么。希望你会喜欢。
Speaker 300:35 - 01:15
Hello, everyone, and welcome to AI and I. This is a slightly different version of the pod today, where we've got myself, Josh, Austin, and Douglas here, and we're missing Daniel. We're gonna be taking you on a journey today, introducing you to All Access in the builder pack. And this is a new tier at Every that we decided to launch recently, and we would just wanna talk about why we decided to build it, why it's important, how we use it, and what it means to be a builder today. All of which all of us are are are builders and use these tools every day.
Speaker 300:35 - 01:15
大家好,欢迎来到 AI and I。今天这期播客会稍微有点不一样,现场有我自己、Josh、Austin 和 Douglas,不过 Daniel 缺席了。今天我们会带大家做一次深入了解,介绍 All Access 和 Builder Pack。这是 Every 最近决定推出的一个新 tier(等级方案),我们想聊聊为什么决定做它、它为什么重要、我们自己怎么使用它,以及在今天这个时代,成为一个 builder(构建者)意味着什么。我们几个人其实都是 builder,而且每天都在使用这些工具。
Speaker 301:15 - 01:41
So, yeah, hopefully by the end of this, there'll be like one nugget that you can take out of this and and and use in in your everyday. So I guess, like, to introduce us or to introduce this this pod, first, I'm gonna introduce All Access. We launched it Tuesday on the fourteenth. It was a big launch for us. Big endeavor across the whole company.
Speaker 301:15 - 01:41
所以,希望到最后,你至少能从这里带走一个有用的点,并且用到你的日常工作里。那我想,先介绍一下我们,或者说先介绍一下这期播客的主题。首先我要介绍的是 All Access。我们在 14 号星期二发布了它。对我们来说,这是一次很大的发布,也是一次涉及全公司的重大项目。
Speaker 301:41 - 02:22
All Access is a new tier. At every it's a $625 annual plan that gives you access to the builder pack, which is 10 benefits, exclusive benefits from all AI companies that you know and love, that enable you to go build. And adopt some of these tools in ways that you may not have been able to before. And All Access includes a couple other things. It includes access to office hours that we're hosting that are that are exclusive to All Access members, unlimited use of Quora and Spiral, and we have a couple other things up our sleeve that we're gonna be launching in the next few months as well.
Speaker 301:41 - 02:22
All Access 是一个新的 tier。在 Every,它是一个 625 美元的年度方案,购买后你可以获得 Builder Pack 的访问权限。Builder Pack 包含 10 项福利,都是来自那些你熟悉且喜爱的 AI 公司提供的专属福利,帮助你真正开始构建,并以你以前可能做不到的方式采用这些工具。All Access 还包含另外几项内容:包括参加我们举办的 office hours(办公答疑时段)的权限,这些是 All Access 会员专属的;还包括对 Quora 和 Spiral 的无限使用;另外,我们还有一些别的内容正在筹备中,也会在接下来的几个月里推出。
Speaker 302:24 - 03:03
But we really wanna talk about the Builder Pack, because that's like probably the meat of of All Access. And the builder pack is 10 soon to be more exclusive benefits to AI tools that the every team uses every day, so that you can get access to these tools and use them, you know, for free for a while and and get comfortable with using them in your stack. It's it's it's early access, exclusive access to the EveryStack. This is the team that that brought it to life. But, Yash, do you use the builder path?
Speaker 302:24 - 03:03
但我们真正想重点聊的是 Builder Pack,因为它大概算是 All Access 最核心的部分。Builder Pack 目前包含 10 项、未来还会继续增加的专属福利,面向 Every 团队每天都在使用的 AI 工具。这样你就可以获得这些工具的使用权限,免费用上一段时间,并逐渐熟悉如何把它们整合进你的 stack(技术栈)里。它可以说是对 EveryStack 的 early access(抢先体验)和 exclusive access(专属访问)。而这就是把它真正做出来的团队。不过,Yash,你会用 builder path 吗?
Speaker 403:03 - 03:09
Yeah. Context. I'm Yash. I run one of the bot like, I'm just Yeah. Yeah.
Speaker 403:03 - 03:09
会。补充一下背景,我是 Yash。我负责其中一个 bot(机器人)……算是,我就是……对,对。
Speaker 303:09 - 03:09
Who are you?
Speaker 303:09 - 03:09
你是谁?
Speaker 403:10 - 03:29
Who who am I? Wow. I ask that myself that every day, but I run one of the products at every called Sparkle, and I also work on the growth team now. A large part of my work is using Cloud. I know at every year, like Codec stands, but I'm still very Cloud.
Speaker 403:10 - 03:29
我是谁来着?哇。这个问题我自己每天都在问,不过我在 Every 负责一个叫 Sparkle 的产品,现在也在 growth team 工作。我的工作很大一部分都在用 Claude。我知道在 Every,像 Codecs 之类的也很重要,但我现在还是非常依赖 Claude。
Speaker 403:29 - 03:55
I've been using Cloud for the last one year. And this involves like, I built out the entire product, like I think 15 rebuilt Sparkle 15 times using Cloud and Posthoc. So those are the tools I reach out for the most. And recently I joined the growth team at Every. And growth meant a lot of AB testing, a lot of segmentation and other stuff.
Speaker 403:29 - 03:55
过去这一年我一直在用 Claude。这包括,比如我把整个产品都搭起来了,我想我大概用 Claude 和 PostHog 把 Sparkle 重建了 15 次。所以这些就是我最常用的工具。最近我加入了 Every 的 growth team。而 growth 意味着大量的 AB testing,以及大量的用户分群和其他事情。
Speaker 403:55 - 04:36
And again, I started optimizing postdoc, but I find it really annoying to open like dashboards and set a lot of these things up. And that's where my setup with Claude helps a lot where I have like custom skills built out. And essentially my plan is in the next few weeks to automate the entire AB testing pipeline. Like a lot of the work that I spent doing for the first one month at the growth team, I'm like, this is very boring because it's it's very straightforward. Like, should be working on interesting more interesting problems and I should not be clicking buttons and figuring out what's the right audience size and stuff like that.
Speaker 403:55 - 04:36
然后,我又开始优化 PostHog,但我发现老是打开各种 dashboard、设置一大堆这类东西,真的很烦。这时候我和 Claude 的那套配置就特别有帮助,我给它做了一些 custom skills。基本上,我的计划是在接下来几周把整条 AB testing 流水线完全自动化。像我在 growth team 第一个月花时间做的很多工作,我都会觉得,这太无聊了,因为它其实非常直接。我们应该去做更有意思的问题,而不是一直点按钮、算什么 audience size 才合适之类的事情。
Speaker 404:36 - 04:48
Like, Claude is so much better at that. I have to prompt it. I have to figure out, okay. What's the right audience size? When should an experiment go from a 10% AB testing cohort to 50%?
Speaker 404:36 - 04:48
比如,Claude 在这方面就强太多了。我只需要 prompt(提示)它,然后搞清楚,好,合适的 audience size 是多少?一个实验应该在什么时候从 10% 的 AB testing cohort(分组)扩大到 50%?
Speaker 404:49 - 05:08
Like, should I be doing that manually? It's I I feel that it's a lot of fake work, and we like to spend a lot of time observing these things, like setting up, like, setting up an experiment every time. It so boring to me. I'm like, the only way I can set up do more experiments is automating the entire pipeline. So I spend a lot of
Speaker 404:49 - 05:08
这种事真的该我手动做吗?我觉得这里面很多都是 fake work(伪工作),而且我们总是花很多时间在观察和搭这些东西上,比如每次都要重新设置一个实验。这对我来说太无聊了。我会想,我能做更多实验的唯一办法,就是把整个流水线自动化。所以我花了很多
Speaker 305:08 - 05:20
time So so what's what's interesting to you is less actually the work. The work is no longer setting up the pipeline to be able to do it. It's actually what should I test? Yeah. And then did it work?
Speaker 305:08 - 05:20
时间——所以,对你来说真正有意思的,其实已经不太是这份工作本身了。工作不再是把流水线搭起来、让这件事能做,而是:我应该测试什么?对。然后它到底有没有效果?
Speaker 305:20 - 05:21
Great. Implement it. What's the next test?
Speaker 305:20 - 05:21
很好。那就实施它。然后下一个测试是什么?
Speaker 405:21 - 05:30
Yeah. I should be like, I'm more interested in the ideas. Like, oh, should we do a homepage takeover? Yeah. Who should we show that homepage takeover to?
Speaker 405:21 - 05:30
对。我应该是那种——我对想法本身更感兴趣。比如,哦,我们要不要做一个 homepage takeover?对。那我们应该把这个 homepage takeover 展示给谁?
Speaker 405:30 - 05:52
Versus actually setting that up is not that interesting. Like, we have so many experiments live right now, which are like, when should we cut off the paywall? When should we show the paywall? Those are way more interesting and those are also like more human psychology. And I can I have more fun doing that?
Speaker 405:30 - 05:52
相比之下,真正把它搭起来这件事就没那么有意思了。比如,我们现在同时在线跑着很多 experiment(实验),像是:我们应该在什么时候切断 paywall(付费墙)?应该在什么时候展示 paywall?这些就有意思得多,而且也更偏向于 human psychology(人类心理)。而且我做这个也会更有乐趣,不是吗?
Speaker 305:52 - 05:52
Mhmm.
Speaker 305:52 - 05:52
嗯。
Speaker 405:52 - 06:08
And and and I personally feel like a large part of our work at Avery is, like, everyone is very secure in what they do, and so we are okay with automating our jobs. Like, I spent one month doing these things, and I'm like, okay. I'm done. Mhmm. Like, I wanna do other things now.
Speaker 405:52 - 06:08
而且,而且,而且我个人觉得,我们在 Avery 的工作有很大一部分是:每个人都对自己做的事情很有安全感,所以我们可以接受把自己的工作自动化。比如,我花了一个月做这些事,然后我就会想,行,可以了。我做完了。嗯哼。比如,我现在想去做别的事情了。
Speaker 406:09 - 06:14
And if it was some other organization, I'm gonna try to stretch this out for, like, a year.
Speaker 406:09 - 06:14
但如果是在别的组织里,我可能会想办法把这件事拖长到一年左右。
Speaker 306:14 - 06:17
Like, what what I did You wanna you wanna move on to the next thing that you tap?
Speaker 306:14 - 06:17
就像,我做的那些——你会想继续到下一个你上手的东西,对吧?
Speaker 406:17 - 06:23
Yeah. Because that's fun, and that's why we are always at the edge of AI. Right? Like, we are like, yeah. This is done, guys.
Speaker 406:17 - 06:23
对。因为那样很有趣,这也是为什么我们总是在 AI 的前沿,对吧?比如,我们会说,行,这个已经做完了,各位。
Speaker 406:23 - 06:27
Like, what's next? Like, what's more interesting? Okay. Is there a better model for this?
Speaker 406:23 - 06:27
那接下来呢?什么更有意思?好,那这件事有没有更好的 model(模型)?
Speaker 306:28 - 06:32
Is So so what in the builder pack Yeah. You use every day?
Speaker 306:28 - 06:32
那么,所以在 builder pack 里——对。你每天都会用什么?
Speaker 406:32 - 06:54
Every day, it's Cursor. I use Cursor because they have cloud regions, which are really good. We did a lot a bunch of Fable five optimizations because I was working on some other task. I did not wanna mess with my setup. So cloud agents just spin off a cloud server and run there and give get get me the results back.
Speaker 406:32 - 06:54
我每天都用的是 Cursor。我用 Cursor 是因为他们有 cloud regions(云区域),这点真的很好。我们之前做了很多 Fable five 的优化,因为当时我在处理别的任务。我不想折腾我现有的 setup(配置环境)。所以 cloud agents(云端 agent)可以直接拉起一台 cloud server(云服务器)在那边运行,然后把结果返回给我。
Speaker 406:55 - 07:22
And they have a really good open source model, which is really cheap, like, unique composer, which I love. So that's something I use. I used Framer a lot as well while I was building the Sparkle site. It it has a it it has good MCP support, I could change the change the headlines, see it, like experiment more directly without actually going into framework and setting. Like, every time it's very annoying to me.
Speaker 406:55 - 07:22
而且他们有一个非常好的 open source model(开源模型),价格很便宜,比如 unique composer,我很喜欢。所以这是我会用的东西。我在做 Sparkle site 的时候也用了很多 Framer。它有很好的 MCP support(MCP 支持),我可以直接改 headline(标题)并立刻看到效果,更直接地做实验,而不用每次都进 framework(框架)里去设置。那对我来说每次都很烦。
Speaker 407:22 - 07:37
Like, I'm setting a lot of these things up the right font size for mobile and stuff like that. Yeah. This is how you like like, I I used to do this, like, twenty years back. Like, today, I should not be spending my time on these things. And I would say Claude Max is my go to.
Speaker 407:22 - 07:37
比如我还得去设置很多这种东西,像 mobile 端合适的 font size(字体大小)之类的。对,这就是你会去做的那种事——我二十年前确实会这样做。但放到今天,我不该把时间花在这些事情上。我会说 Claude Max 是我的首选。
Speaker 407:37 - 07:42
I Yeah. Like, I I feel it's insane. Yeah. That's my stack. It's tough.
Speaker 407:37 - 07:42
我……对,我觉得这简直离谱。对,这就是我的 stack(技术栈)。挺难的。
Speaker 407:42 - 07:51
Yeah. And post hoc. Yeah. I use a lot of these products, but it's so cheap. Like, I feel that Cloudmax subscription is really underrated.
Speaker 407:42 - 07:51
对,还有 post hoc。对,我用了很多这类产品,但它其实非常便宜。比如我觉得 Cloudmax subscription(Cloudmax 订阅)真的被严重低估了。
Speaker 407:51 - 07:54
Like, I'd pay way more for it. Right. I couldn't work without it.
Speaker 407:51 - 07:54
我愿意为它付更多钱。真的,没有它我根本没法工作。
Speaker 307:54 - 07:58
Right. Anthropic. I hope no one from Anthropic is watching this.
Speaker 307:54 - 07:58
对,Anthropic。希望没有 Anthropic 的人在看这个。
Speaker 407:58 - 08:01
Yeah. Get the d, guys. Get get get get the Builder pack.
Speaker 407:58 - 08:01
对,去买 d,伙计们。去买买买买那个 Builder pack。
Speaker 508:02 - 08:45
I I like the way Yasha is describing, one potential approach to work, which is like, there there's this there's this approach to being a general manager in the NBA, which is like, just keep trading for future draft picks to be like, we're on an eight year time horizon for contention. And there is for someone as, like, skilled as Josh at engineering, go to market engineering growth, you can use these tools, this stack that he has to to, like, operate at the same level and, like, a set it and forget it mode and kind of I would be like, well, Yash like, I I run the growth team. I look at what Yash is doing. I'd be like, Yash is another experiment, and it, like, moves the needle by 2%. I guess he's on it.
Speaker 508:02 - 08:45
我喜欢 Yasha 描述的一种潜在工作方式,有点像 NBA 里当 general manager 的一种思路:不断交易、换取未来的 draft picks,就像在说“我们争冠的时间线是八年”。而对于像 Josh 这样在 engineering、go to market、engineering growth 方面都非常有能力的人来说,你可以用他手上的这些工具、这套 stack(技术栈),以同样的水准运行工作,进入一种 set it and forget it(设好后基本自动运行)的模式。某种程度上我会想,如果我是 Yash——我是带 growth team 的——我看 Yash 在做什么时,可能会觉得:“Yash 又是另一个 experiment(实验),它大概把指标推进了 2%,行吧,那他应该在持续跑这件事。”
Speaker 508:45 - 09:33
And, like, he knows how to set up that that Claude code and post hoc loop. But the way that he works instead, which is exciting, is that after he does that, he looks at how can we win right now? How can I be creative and competitive to find the next thing? And how can I use these always on, like, loop and architecture based automations to make it so that Yash can do the thing that, like, lights them up, which is I can see Yash in Slack being like, oh, what if we, like, made a three hour time constrained discount specifically targeted to these users using this video, building this thing in Framer? And that's, this kind of stuff has always been the case, right?
Speaker 508:45 - 09:33
而且,他也知道怎么搭起 Claude code 和 post hoc loop。但真正令人兴奋的是,他实际的工作方式不是停在那里,而是在做完这些之后,会继续想:我们现在怎么赢?我怎样才能更有创意、更有竞争力地找到下一个机会?我又怎样利用这些 always on(常开)、基于 loop 和 architecture(架构)的自动化,让 Yash 可以去做那些真正让他兴奋的事?我甚至都能想象 Yash 在 Slack 里说:“如果我们做一个三小时限时折扣,专门针对这些用户,配上这个视频,再用 Framer 搭这个东西,会怎么样?”而这种事其实一直都是这样,对吧?
Speaker 509:33 - 10:22
Like that work is the most fun way to work. It's how smart, creative people are able to differentiate themselves. But I think that the builder pack allows is that it allows someone like Josh to automate and defer a lot of the work to cloud agents. He would have otherwise had to manually click in post hoc or manually code himself, and then he can go be creative on the space where he is an expertise. And then it also allows someone like me who doesn't know what the fuck any of that stuff is, but kind of is naturally more of an ideas person to give the model and the SaaS tools in the builder pack my outcomes and ideas, and then just, like, let it run, which is, like, the craziest part for me.
Speaker 509:33 - 10:22
这种工作方式本来就是最有趣的工作方式,也是聪明、有创造力的人真正拉开差距的方法。但我觉得 builder pack 带来的,是它让像 Josh 这样的人可以把大量工作自动化,并且交给 cloud agents(云端 agent)去做——这些工作本来要么得他自己手动在 post hoc 里点,要么得他亲自写代码。这样一来,他就可以把精力放到自己真正擅长、能够发挥专业能力的创造性空间里。与此同时,它也让像我这样的人——我根本搞不懂那一堆东西到底是什么鬼,但天然更偏 ideas person(点子型的人)——可以把我的结果目标和想法交给 model(模型)以及 builder pack 里的 SaaS tools(SaaS 工具),然后就让它自己跑起来。对我来说,这才是最疯狂的部分。
Speaker 510:22 - 10:25
Like, I I let go oh, yeah.
Speaker 510:22 - 10:25
就是,我……我放手让它去做了。哦,对。
Speaker 310:25 - 10:30
Can you tell that story of, you had an idea for something? You went to bed. You woke up. You had four emails.
Speaker 310:25 - 10:30
你能讲讲那个故事吗?就是你有了一个想法,然后去睡觉,醒来之后收到了四封 email。
Speaker 510:31 - 10:43
So it wasn't even my idea. Like, we were we did the we did the builder pack launch on Tuesday. It was going really well. I'm on the West Coast. Most of the team is on the East Coast.
Speaker 510:31 - 10:43
其实那甚至都不是我的想法。我们周二做了 builder pack 的发布,进展非常顺利。我人在 West Coast,团队大部分人在 East Coast。
Speaker 510:43 - 11:26
As they were, like, getting ready to go to bed, Josh dropped a note in Slack saying, hey, we should send emails in the morning to a group of users who have shown some sort of checkout intent but not converted because we were running a early bird discount for the first twenty four hours of launch. So it's like we should probably hit them again with some sort of urgency backed approach. And then Yosh said, this isn't actually my idea, it's Dan's idea, but I'm sharing it. And then they all signed off. And I think maybe like 7PM here, I had been in front of my computer all day, and I really wanted to go to the gym.
Speaker 510:43 - 11:26
当时他们差不多准备睡觉了,Josh 在 Slack 里留了一条消息,说,嘿,我们明早应该给一批用户发 email——这些用户已经表现出某种 checkout intent(结账意图),但还没有转化,因为我们在发布后的前二十四小时做了 early bird discount(早鸟折扣)。所以感觉我们应该再触达他们一次,用某种更有紧迫感的方式来推进。然后 Yosh 说,这其实不是我的主意,是 Dan 的主意,但我来转发一下。接着他们就都下线了。我这边大概晚上 7 点,我已经一整天都坐在电脑前了,特别想去健身房。
Speaker 511:26 - 11:46
And so I I took a screenshot of the Slack because it was like a bunch of different Slack messages. Usually, would just like copy paste the link to the Slack message and send it to Codex. I took a screenshot of the Slack. I dropped it in a Codex thread using five six on Medium. It's like usually what I operate in.
Speaker 511:26 - 11:46
所以我就把 Slack 截了个图,因为那里面其实是好几条不同的 Slack 消息。平时我一般会直接复制粘贴那条 Slack 消息的链接,然后发给 Codex。这次我截了 Slack 的图,把它丢进一个 Codex thread(线程)里,用的是 five six on Medium——这基本算是我平时最常用的配置。
Speaker 511:46 - 12:36
And I said, can you do this? And then I did slash LFG, which is the harness in our compound engineering plugin to tell an agent to take itself through a brainstorm, plan, work, review, compound loop, which is like, to me, the ideal way to work with with agents. And then I went to the gym, and I came back from the gym, and what I saw in Codex was that it works for honestly not that long. Like, it it probably finished by the time that I, like, changed into my workout clothes at the gym because it, like, didn't actually need to to do that much. And it had created four different audience based cohorts, like targeted cohorts in kits based on, like, these people got to the checkout page.
Speaker 511:46 - 12:36
我就说,你能做这个吗?然后我用了 slash LFG,也就是我们 compound engineering plugin 里的那个 harness,用来告诉一个 agent 让它自己经历一轮 brainstorm、plan、work、review、compound loop;对我来说,这基本就是和 agents 协作的理想方式。然后我去健身了,健身回来后,我在 Codex 里看到的是,它其实没花那么久。大概我刚到健身房换好运动服的时候,它可能就已经做完了,因为这件事本身其实也不需要做太多。它已经创建了四个不同的、基于 audience 的 cohort,也就是 kit 里那种定向 cohort,比如说,这些人曾经到过 checkout page。
Speaker 512:36 - 13:00
These are, these people visited the webs visited the BuilderPack website but didn't check out. These are our highest engaged free users. There's, like, four different types. It drafted specific emails for each of those audience types. It ran the text for the emails through the MCP for our AI based writing app called Spiral.
Speaker 512:36 - 13:00
这些人访问过 BuilderPack website,但没有结账。这些是我们参与度最高的免费用户。大概有四种不同类型。它还为每一种 audience 类型分别起草了对应的 email。它把这些 email 的文本通过 MCP 送进了我们那款基于 AI 的写作 app——Spiral。
Speaker 513:00 - 13:27
So it's using a writing style that we've approved. It looked through all of our previous sends in kit, both for our language style, as well as what had driven click throughs, had driven opens. It grabbed a social share image that had driven click throughs in a previous send and embedded it. And it had scheduled it for the next morning. And it's crazy because I wasn't like, holy shit, that's so cool.
Speaker 513:00 - 13:27
所以它用的是我们已经批准过的写作风格。它查看了我们之前在 kit 里的所有发送记录,既分析了我们的语言风格,也分析了哪些内容带来了 click-through、哪些带来了 open。它还抓取了一张在之前某次发送中带来过 click-through 的 social share image,并把它嵌进去了。而且它还把发送安排到了第二天早上。疯狂的是,我当时并不是那种“holy shit,太酷了”的反应。
Speaker 513:27 - 13:34
I was like, that's basically what it should do. Like, my brain is so cooked on this stuff now that I'm like, this is literally exactly what I expected.
Speaker 513:27 - 13:34
我的反应更像是,这基本就是它本来就该做的事。我的大脑现在已经被这些东西“腌透”了,以至于我会觉得,这完全就是我预期它应该做到的样子。
Speaker 413:34 - 13:35
Did
Speaker 413:34 - 13:35
它有
Speaker 313:35 - 13:39
something I think it then posted it in all emails.
Speaker 313:35 - 13:39
我想它后来还把这件事发到了 all emails 里。
Speaker 513:40 - 14:12
Yeah. I have I have an automation in there that it like so we have we have a Slack channel called all emails. If you're a subscriber to every, I'm I'm sorry, we send you a lot of emails. I swear we're working on it. One one step one step towards working on it is having a channel where one, we we build all of our email sends through Codex or Cloud Code and can have the agent push a Slack message saying, here's what here's what's gonna be sent to whom and why.
Speaker 513:40 - 14:12
对。我在里面配了一个 automation,所以它会——我们有一个 Slack channel 叫 all emails。如果你订阅了我们所有内容,呃,我很抱歉,我们确实给你发了很多 email。我发誓我们正在努力解决这个问题。我们朝着解决它迈出的一步就是建了这样一个 channel:第一,我们通过 Codex 或 Cloud Code 来构建所有 email sends;第二,我们可以让 agent 推送一条 Slack message,说明接下来要给谁发送什么、以及为什么这么发。
Speaker 514:12 - 14:51
And then they can get a review of, you know, should we actually send this to this audience at this time? So it sends that as well, and the team can review it and know and know what's up. I but I I pulled up I pulled up these four I pulled up these four email sends, and I made very light tweaks, partially because I was like, oh, I think I can write, like, a slightly better headline and also partially because I'm still a control freak who doesn't want to turn over every single thing to AI. I was like, let me, you know, let me, like, put a little bit of spice on this, but I I maybe spent ten minutes of extra work on it when I came back from the gym. And then we we scheduled them and we we let it rip.
Speaker 514:12 - 14:51
然后大家就可以审查一下,比如说,我们到底该不该在这个时间把这封邮件发给这批 audience?所以它也会把这些发出来,团队就能 review,也能知道当前是什么情况。不过我当时调出了这四封 email sends,只做了非常轻微的修改,一部分是因为我想,哦,我也许能写一个稍微更好的 headline;另一部分也是因为我骨子里还是个 control freak,不想把每一件事都完全交给 AI。我当时想的是,让我稍微给它加一点 spice 吧。但我从健身房回来后,额外花在这上面的时间可能也就十分钟。然后我们就把它们排好发送时间,直接放手让它跑了。
Speaker 514:51 - 15:16
And in in a couple hours the next morning, those emails alone had driven more than $25,000 in revenue. It had driven, like, multiple conversions. It suggested things we could do to retarget the audience that didn't convert. And, like, two nice things happened. Like, one, I got to go to the gym, which is great, because otherwise, would have sat there building these these emails.
Speaker 514:51 - 15:16
到第二天早上再过几个小时,光是那些邮件就带来了超过 $25,000 的收入。它带来了好几次转化(conversion)。它还建议了我们可以做哪些事情,去重新触达那些没有转化的受众。而且,发生了两件很棒的事。第一,我可以去健身房了,这非常好,因为不然的话,我本来会坐在那里手动写这些邮件。
Speaker 515:17 - 15:47
And two, like, at the gym, not looking at my phone, not looking at Slack, I had other ideas for stuff we could do the next day. Right? Like, disconnected from being in the weed, I'm, like, working out, and I'm like, oh, yeah. We should, like, we should do this experiment and that experiment because I don't have to use my brain and my fingers to, like, click around in postdoc, which I which I love using, but I don't have to click around in postdoc to, like, find the exact right audience segment to to target.
Speaker 515:17 - 15:47
第二,在健身房里,不看手机,不看 Slack,我反而想到了一些我们第二天还可以做的事情。对吧?就是,当我从那些繁琐细节里抽离出来时,我一边锻炼,一边就在想,哦,对了,我们应该做这个实验、那个实验,因为我不用把脑子和手指花在 postdoc 里到处点来点去——我其实很喜欢用它——但我不需要再在 postdoc 里不停点击,只为了找到那个完全正确、值得投放的受众细分。
Speaker 315:47 - 16:37
That that's it's just so crazy, like, to go from here's an idea to go do this idea using slash LFG to actually, like, executing on what it created, in this case, sending these emails, to generating $25,000 in revenue, which, like, doesn't happen every time you do last LFG, but, like, that is, you know, sort of a possibility to then compounding what you learned and also recommending, like, what to do next. It's just, like, remarkable that, you know, like, 5.6 SOL is expensive. But, like, relative to the value that you got out of that, it's, like, nothing. And best of all, you went to the gym and like took care of yourself. And that's it's just like remarkable that you can do that.
Speaker 315:47 - 16:37
这真的太疯狂了:从“这里有一个想法”,到用 /LFG 去执行这个想法,再到真正落实它所生成的内容——在这个例子里,就是发出这些邮件——然后带来 $25,000 的收入。虽然这并不是每次用 /LFG 都会发生,但这确实是一种可能性。而且你还可以把学到的东西持续叠加,并获得关于下一步该做什么的建议。这真的很了不起。没错,5.6 SOL 很贵,但和你从中获得的价值相比,根本不算什么。最棒的是,你还去了健身房,照顾了自己。你居然能做到这一点,这本身就非常惊人。
Speaker 316:37 - 16:51
You can if you're really harnessing these tools well for work and for life, I use it as my trainer at the gym, It's it can be very fulfilling and and and, like, so effective.
Speaker 316:37 - 16:51
如果你真的能很好地把这些工具用于工作和生活——我甚至把它当作我在健身房的 trainer(教练)——那它会让人很有满足感,而且真的、真的非常有效。
Speaker 416:52 - 17:09
Yeah. I feel it allows you to be get better at things you wanna really get better at. It frees up more time. To give an example, when Austin and I people are always pitching growth ideas to us. Like, hey, do this, do this.
Speaker 416:52 - 17:09
对,我觉得它能让你把更多精力放在那些你真正想变得更擅长的事情上。它释放了更多时间。举个例子,Austin 和我总是会收到别人给我们提增长(growth)点子的建议。比如,嘿,做这个,做那个。
Speaker 417:09 - 17:32
And to be really honest, those are good ideas, but I don't want to take away time from doing my main thing and spend time on that. And that entire thing should be automated because, a, the idea is good. We should not hire an engineer just to implement that particular idea. So I feel we have these tools as like if I have a workflow set up, hey. Let's try this particular headline on our website.
Speaker 417:09 - 17:32
说实话,那些主意其实都不错,但我不想从我的核心工作里挤出时间,再花时间去做那些事。而整件事本来就应该自动化,因为,第一,这个想法是好的;第二,我们不应该为了实现某一个具体想法,就专门去雇一个 engineer(工程师)。所以我觉得我们有了这些工具之后,如果我已经搭好了一个 workflow(工作流),比如说,嘿,在我们的网站上试试这个标题。
Speaker 417:32 - 17:40
It should be like, you should be that should be automated. Yep. And I feel we can that's the goal. Like, I have more time to do, like, deep work
Speaker 417:32 - 17:40
那这件事就应该是自动化的。对。我觉得我们的目标就是做到这一点。这样我就有更多时间去做那种真正深入的工作,
Speaker 317:40 - 17:40
Yeah.
Speaker 317:40 - 17:40
对。
Speaker 417:40 - 17:49
And or and be in my flow state and without really compromising these things. Like, building that loop is me being in the flow state Yeah. Versus, like, hey. Try this. Try this.
Speaker 417:40 - 17:49
或者说,进入我的 flow state(心流状态),而不用真的为了这些事情做出妥协。比如,构建那个循环本身,就是我处在心流状态的一种表现。对。相比之下,不是那种“嘿,试试这个,试试那个”。
Speaker 417:49 - 17:50
Try this.
Speaker 417:49 - 17:50
试试这个。
Speaker 517:50 - 18:46
And I I think the thing we're we're, like, circling around is, this is the best way to to build right now, which is to, you know, Kieran, our colleague, has described it this way. You're at the top and bottom of the AI sandwich. You you you decide what the problem is, you frame the idea, and then you review the work. That's been the case for a while. But the exciting thing about this moment, the exciting thing about being able to use the tools in the Builder Pack is that you can go from idea to execution quickly, easily, reliably, and at like a very high level by being able to just like, type or with monologue, any speech to text tool like we use just to be able to say the idea and have it come out.
Speaker 517:50 - 18:46
而且我觉得,我们一直在反复靠近的那个点是:这就是当下最好的构建方式。就是,Kieran,我们的同事,是这么描述的:你处在 AI 三明治的顶部和底部。你来决定问题是什么,你来框定这个想法,然后你来审核结果。这种情况其实已经持续一段时间了。但此刻真正令人兴奋的地方,真正让人兴奋的是能够使用 Builder Pack 里的工具,让你可以快速、轻松、可靠地、而且以很高的水准,把一个想法直接变成执行成果;你基本上只需要打字,或者用 monologue、以及任何像我们在用的 speech to text(语音转文字)工具那样的工具,把想法说出来,它就能产出结果。
Speaker 518:46 - 19:15
And the thing that I really like about this, the way I like working is that whether you're, you know, primarily a Claude user like Yash or a Codex user like me, the tools all now seek together really, really well and connect to each other really well. Notion is is a great example of this. I have faded Notion for years because my brain doesn't work in that way of like, I'm gonna like organize this database and I'm gonna make sure that like the growth
Speaker 518:46 - 19:15
而我真的很喜欢这一点。我喜欢的工作方式是:无论你主要是像 Yash 那样用 Claude,还是像我这样用 Codex,现在这些工具彼此之间都能配合得非常、非常好,连接也非常顺畅。Notion 就是一个很好的例子。多年来我一直不太用 Notion,因为我的大脑不是那种运作方式,不会想着“我要把这个 database(数据库)整理好,我要确保这个 growth
Speaker 319:15 - 19:17
hub works. It works.
Speaker 319:15 - 19:17
hub 能运转”。它确实能运转。
Speaker 519:17 - 19:41
Yeah. Yeah. Like, I think up until a year ago, Notion was had had too much of a like, of a barrier to entry on like, organization and main and maintenance that anytime I try to use Notion with a with a team, we've churned because no one's around to maintain it. And all of that maintenance has now been pushed off to Codex and Cloud Code. Once a week, I just asked Codex and I'm like, Hey, can you like reorganize our growth hub?
Speaker 519:17 - 19:41
对,对。我觉得直到一年前,Notion 在组织、管理和维护这件事上,门槛还是有点太高了。以前我只要尝试和团队一起用 Notion,最后往往都会流失,因为没有人会一直在那儿维护它。而现在,这些维护工作基本都转交给 Codex 和 Cloud Code 了。现在我每周只要问 Codex 一次:嘿,你能帮我们把 growth hub 重新整理一下吗?
Speaker 519:41 - 20:03
And it does it perfectly. It's like amazing. Or it's like, hey, make sure this database is set up well. My favorite example of this is while I was doing the this kind of like, in the weeds Codex work, like when I work with Codex, I'm like steering it and going back and forth with it. Claude in in Claude code in the desktop app using Fable.
Speaker 519:41 - 20:03
然后它就能做得非常完美,真的很惊人。或者我会说,嘿,帮我确认这个 database 设置得够不够好。我最喜欢的一个例子是:当时我在做这种非常深入细节的 Codex 工作时——我和 Codex 协作的时候,基本上就是我在引导它、和它来回交互——与此同时,桌面 app 里的 Claude,在 Claude code 中配合 Fable
Speaker 520:03 - 20:54
I've had the videos we're releasing attached to this launch all connected through Descript in the Descript MCP. And Sable is just running in a loop, building out the edits for all of those videos and reviewing its work and and interacting with Underlord, the Descript AI tool, because I'm not a video editor. Like, I'm actually really bad at it, but, like, sometimes that falls under my responsibilities. And that can run-in a loop so that when I go to edit the video, by no means has has Fable and Descript created a video that is publishable, but it has gotten me like 70% of the way there while I'm steering Codecs, Cloud Code and Fable and and Descript have gotten it there so that I can go and I'm like, oh, I know what to do now. Like, I see the pieces.
Speaker 520:03 - 20:54
一起工作。我把这次发布要上线的视频,全都通过 Descript 和 Descript MCP 连接起来了。然后 Sable 就一直在循环运行,持续为所有这些视频搭建剪辑、审核它自己的结果,并且和 Underlord——Descript 的 AI 工具——交互。因为我不是视频剪辑师,事实上我在这方面挺不擅长的,但有时候这又确实属于我的职责范围。所以它可以一直这样循环运行。等到我真正去编辑视频时,Fable 和 Descript 当然远远还没有做出一个可以直接发布的视频,但它们已经帮我完成了大概 70% 的工作。而在我引导 Codex 的同时,Cloud Code、Fable 和 Descript 已经把事情推进到了那个程度,让我一过去就会觉得:哦,我现在知道该怎么做了。我已经能看清这些组成部分了。
Speaker 520:54 - 21:07
It like, it does this crazy stuff where it takes the transcript, makes a storyboard, makes a script itself, and then, like, pieces the video together. So I just have to do, like, a few tiny things to get us to a final product.
Speaker 520:54 - 21:07
它会做这种很疯狂的事:拿到 transcript(转录文本)之后,先做 storyboard(分镜板),再自己生成 script(脚本),然后把视频一点点拼接起来。这样我只需要再做几个很小的处理,就能把它变成最终成品。
Speaker 221:07 - 21:34
You already know how AI is changing how everyday work gets done, how much ground you can cover, and how fast a team can scale. To stay ahead, you need tools that give you a competitive advantage built for this new era. Adio is the CRM for the agent native world. It meets you where you work, compounds every customer signal into context, and then acts on it across your pipeline to let you move at unmatched speed and scale. With agents and automations for every job, Adio orchestrates your work around the clock.
Speaker 221:07 - 21:34
你已经知道 AI 正在如何改变日常工作的完成方式、你能覆盖多大的范围,以及一个团队能以多快的速度扩张。想保持领先,你需要的是能为这个新时代打造竞争优势的工具。Adio 是面向 agent native 世界的 CRM。它会在你工作的地方与你衔接,把每一个客户信号不断累积成上下文,然后在你的整个销售管道中据此采取行动,让你以无与伦比的速度和规模推进工作。借助适用于各种岗位的 agents 和自动化,Adio 全天候编排你的工作。
Speaker 221:34 - 21:52
We use it internally at every, and we love it. It's built to handle the scale of your workloads. It's extensible with an API and MCP access, and is built with infrastructure to keep up with your most ambitious agents. It's loved by high growth startups like Granola, Modal, WhisperFlow, and Every. Adio runs the work behind Everywin.
Speaker 221:34 - 21:52
我们在 Every 内部到处都在用它,而且我们很喜欢。它被构建出来就是为了处理你工作负载的规模。它可以通过 API 和 MCP 访问来扩展,并且底层基础设施的设计就是为了跟上你那些最有野心的 agent(智能代理)。像 Granola、Modal、WhisperFlow 和 Every 这样的高速增长 startup 都很喜欢它。Adio 支撑着 Everywin 背后的工作。
Speaker 221:53 - 22:03
That's Adio, the agentic serum. Go to adio.com/every and get 15% off your first year. That's adio.com/every. And now, back to the episode.
Speaker 221:53 - 22:03
那就是 Adio,the agentic serum。前往 adio.com/every,你第一年的费用可享受 15% 折扣。网址是 adio.com/every。现在,回到本期节目。
Speaker 322:03 - 22:17
I think the thing that I love about the builder pack specifically is I feel like a lot of the time at Everywhere telling people like, go build. Go build. It's the best time in the world world to build. There's no excuses. Be the builder you wanna be.
Speaker 322:03 - 22:17
我觉得我特别喜欢 builder pack 的一点是,很多时候我们在 Everywhere 都在告诉大家:去构建。去构建吧。现在是这个世界上最适合构建的时代。没有借口。去成为你想成为的 builder(构建者)。
Speaker 322:17 - 22:37
And that's great. But how? It's a it well, one is how, which is like, you should subscribe to every, we will teach you how. But two is, this stuff is expensive. And the stack that you use, Austin, the stack that you use, Josh, Douglas, myself, is very expensive.
Speaker 322:17 - 22:37
这些话当然很好。但问题是,怎么做?一方面是“怎么做”的问题,也就是,你应该订阅 Every,我们会教你怎么做。另一方面是,这些东西很贵。而且你在用的 stack(技术栈),Austin 你在用的 stack,Josh、Douglas,还有我自己在用的 stack,都非常昂贵。
Speaker 322:37 - 23:03
We spend a lot of money at every on AI and credits. And last night, we had an event in the office, and a friend of mine's younger brother, I invited him to come. He's an engineer, and he he's trying to figure out how to get a job as an engineer right now. Even though he is an engineer, he's really struggling to get a job, And he's not using AI at all. Yeah.
Speaker 322:37 - 23:03
我们在 Every 上花了很多钱在 AI 和 credits(额度)上。昨天晚上,我们在办公室办了个活动,我邀请了一个朋友的弟弟过来。他是一名 engineer(工程师),现在正想办法找一份工程师的工作。尽管他本身就是工程师,但他现在找工作真的很困难,而且他完全没有在用 AI。对。
Speaker 323:03 - 23:35
And I introduced him to some people that were like telling them how they work. And then at the end of it, he was like super excited. And then he these people were telling them like how much money their companies enable them to spend on AI. Some guy, you know, he's like a single engineer who spent $30 last month. And this guy like, you know, I could see him like visually like gulp while he was like, how am I going to discover this technology when it's like prohibitively expensive as an and I'm an individual.
Speaker 323:03 - 23:35
我把他介绍给了一些人,那些人跟他讲了讲他们是怎么工作的。到最后,他变得特别兴奋。然后这些人又告诉他,他们所在的公司愿意让他们在 AI 上花多少钱。比如有个人,你知道的,他就是一个单独工作的工程师,上个月花了 30 美元。然后我能明显看出来,这家伙像是当场咽了口口水,心里在想:当这项技术贵得让人难以承受,而我又只是个个人用户时,我到底要怎么接触并掌握它?
Speaker 323:35 - 24:15
I'm not at a company right now. And the thing that I love about the builder pack is it is, you know, I think the the codec, the OpenAI benefit is like a great example of this. Because you're getting three months free of OpenAI and codecs, and a thousand dollars in overage credits. But the three months free includes, I don't know what the number is anymore, $5,000 of value. Like, these these subscriptions just like our our, you know, our Anthropic benefit, they include so much spend inside of a small subscription.
Speaker 323:35 - 24:15
我现在又不在一家公司里。而我喜欢 builder pack 的地方就在于,它——我觉得 OpenAI 的 benefit(权益)就是个很好的例子。因为你能拿到 OpenAI 和 Codex 三个月免费使用,以及 1000 美元的超额 credits(额度)。但这三个月免费里包含的价值,我现在也不确定具体数字了,大概有 5000 美元的价值。像这些订阅,就像我们的 Anthropic benefit 一样,在一个小额订阅里面其实包含了大量可用额度。
Speaker 324:17 - 24:33
So, I mean, in in it's the same thing for basically OpenAI, Cursor, Gemini, and Anthropic, those offerings, because they're they're model companies. So okay. It's one month. Excuse me. The one month offer.
Speaker 324:17 - 24:33
所以,我的意思是,OpenAI、Cursor、Gemini 和 Anthropic 基本上也都是同样的情况,这些 offering(套餐)都是这样,因为它们都是 model companies(模型公司)。所以,好吧。是一个月。抱歉,我说错了。是这个一个月的优惠。
Speaker 324:33 - 24:33
You
Speaker 324:33 - 24:33
你
Speaker 424:34 - 24:35
can use the Codex credit.
Speaker 424:34 - 24:35
可以使用 Codex credit。
Speaker 324:35 - 24:38
You can use the Codex credit for three months. Yeah.
Speaker 324:35 - 24:38
你可以使用 Codex credit 三个月。对。
Speaker 424:38 - 24:38
Yeah.
Speaker 424:38 - 24:38
对。
Speaker 324:38 - 25:01
Alright. It's so much value that, like, you then get to go experiment with the technology so that you can Yeah. Be the builder that you want to be. I use them all the time. My my stack I mean, the the builder pack is basically everything that we use at every personally, I'm in OpenAI, Anthropic.
Speaker 324:38 - 25:01
好。它的价值非常大,像是,你随后就可以去试验这项技术,这样你就能——对——成为你想成为的 builder(构建者)。我一直都在用它们。我的 stack(技术栈),我是说,builder pack 基本上就是我们在每一处都会用到的东西;就我个人而言,我会用 OpenAI、Anthropic。
Speaker 325:01 - 25:34
I'm more of like a consumer of Posthog than I am like a user of of like their AI credits. You're always in Notion. I'm a big Notion user because I do, like, a lot of ops, framer user, render, and I really like experimenting with the Flora MCP. So, I mean, I'm kind of all over the stack. My my biggest use is like, obviously, like the model companies and just like doing my work inside of inside of cloud or codecs.
Speaker 325:01 - 25:34
相比真正去用 Posthog 的 AI credits,我更像是 Posthog 的普通使用者。你总是在 Notion 里。我是 Notion 的重度用户,因为我会做很多 ops(运营)工作,也用 framer、render,而且我也很喜欢试验 Flora MCP。所以,我基本上整个 stack 都会碰。对我来说,最大的一类使用场景显然还是 model companies(模型公司),以及就在 cloud 或 codecs 里面完成我的工作。
Speaker 325:35 - 25:52
I mean, ironically, a lot of these tools, almost all of these tools, if they aren't a model company themselves, I'm utilizing them through a model company. I'm never really ever going into their tool. Even Flora, which like has like a remarkable tool, I only ever use it through the MCP.
Speaker 325:35 - 25:52
我的意思是,讽刺的是,这里面很多工具,几乎所有这些工具,如果它们自己不是 model company,那我就是通过 model company 来使用它们。我实际上几乎从不真正进入它们自己的工具里。即便是 Flora,虽然它本身就是个非常出色的工具,我也只会通过 MCP 来用它。
Speaker 525:53 - 26:24
Yeah. And it doesn't it doesn't devalue those tools. It actually like to me, it incentivizes the those, like, SaaS operators to be the very best at the thing they do. I think a lot of us went through a moment from like January to March where we were like, let's go, we're building our own CRM, we're building our own like, data dashboards, we're gonna do design all ourselves. And you start doing it and you're like, man, this fucking sucks.
Speaker 525:53 - 26:24
对。而且这并不会降低那些工具的价值。实际上,在我看来,这反而会激励那些 SaaS operators(SaaS 运营方)把自己所做的那件事做到最好。我觉得我们很多人大概都经历过一个阶段,从一月到三月,我们会想:来吧,我们自己做 CRM,我们自己做 data dashboards(数据仪表板),设计也都自己搞。然后你真正开始做的时候,就会想,妈的,这也太难受了。
Speaker 526:24 - 26:27
Right? And when you're able to sit
Speaker 526:24 - 26:27
对吧?而当你能够坐下来时
Speaker 326:27 - 26:27
there and be like, you
Speaker 326:27 - 26:27
在那儿,然后就像,你知道的,
Speaker 526:27 - 26:54
know what, I really trust Postdoc to both run and automate my analytics. I really trust Notion to have this source of truth for the entire business. And even if I never open it up, even if my pages and my databases look incomprehensible to a human, it doesn't really matter because when I'm sitting in Cloud Code or Codex, I can which I which I do all the time. I'm like, hey. I need to make this plan.
Speaker 526:27 - 26:54
你知道吗,我真的很信任 Postdoc,既让它运行我的 analytics,也让它把这些流程自动化。我也真的很信任 Notion,让它作为整个业务的 source of truth(单一事实来源)。即使我从来不打开它,即使我的页面和数据库看起来人类根本看不懂,其实也没那么重要,因为当我坐在 Cloud Code 或 Codex 里时,我可以——而且我一直都是这么做的——我就会说,嘿,我需要做这个计划。
Speaker 526:54 - 27:47
I need to do this thing at Notion, at Slack, at Posthog, find the necessary information. That's one reason why the, this suite of tools is so powerful right now is that the leap from 5.5 to 5.6 came with this improvement in the model's ability to go find the context that needs to do work well, and then to maintain that context through any amount compaction that happens as it works over a long term. That happens through the way in which if you looked at my prompts in in Codex, you would be like, I think this guy doesn't know what the hell he's doing. Because I'm I'm usually just like, hey, look at look at all this stuff, create a goal for yourself, and use compound engineering. And I'm like, go.
Speaker 526:54 - 27:47
我需要在 Notion、Slack、Posthog 里做这件事,去找到必要的信息。这也是为什么这套工具现在这么强大的原因之一:从 5.5 到 5.6 的跃迁,带来了 model(模型)在寻找完成工作所需上下文方面能力的提升,而且它还能在长期工作过程中,无论发生多少 compaction(压缩/上下文压缩),都把这种上下文维持住。这一点也体现在一种工作方式上:如果你去看我在 Codex 里的 prompts(提示词),你可能会觉得,这家伙根本不知道自己在干嘛。因为我通常就是这样:嘿,看看这一堆东西,给你自己设定一个目标,然后用 compound engineering。然后我就说,去吧。
Speaker 527:47 - 28:07
It's like barely any thought or work. The thought or work comes from like, I've identified this as the thing to do. And I know that either I or my team have set up the harness that the agents can work inside of to be trusted. And I think to a lot of people that sounds like tough. It sounds like a lot of work.
Speaker 527:47 - 28:07
这几乎不需要什么思考或劳动。真正的思考或劳动更像是:我已经识别出这就是该做的事。而且我知道,我或者我的团队已经搭好了一个 harness(约束/运行框架),让这些 agents(智能体)能在里面工作,并且值得被信任。我觉得对很多人来说,这听起来很难,像是工作量很大。
Speaker 528:07 - 28:32
But the reason that that we've gotten there from Every is we've done it incrementally, right? It's like we did one thing and it started to work okay. We've been, I've been trying to automate these marketing email sends for months, and it keeps getting a little bit better. And through compound engineering, the results keep compounding. So the project I work inside of keeps getting a better understanding of it.
Speaker 528:07 - 28:32
但我们之所以能从 Every 走到这一步,是因为我们是渐进式做到的,对吧?就是先做一件事,它开始运转得还不错。几个月来,我一直在尝试把这些营销邮件发送自动化,而它也一直在一点点变好。通过 compound engineering,结果也在持续复利式累积。所以我所处的这个项目,也在不断对它自身形成更好的理解。
Speaker 528:33 - 28:48
And it kind of goes back to like, the the best time to start doing this was, you know, a year ago. The second best time is right now. You can start this compounding flow right now with models that have have really caught up to how how quickly you can work yourself.
Speaker 528:33 - 28:48
这有点回到那句老话:开始做这件事的最佳时间,是一年前;第二好的时间,就是现在。你现在就可以开始这种复利式流程,而且现在的 models(模型)已经真正追上了你自己工作的速度。
Speaker 428:49 - 29:02
Yeah. I'm curious to know, Douglas, what's your stack? Because you're the creative director behind this. And someone asked me, like, what this branding is very different from every's branding. So, like, why is it green?
Speaker 428:49 - 29:02
对。我很好奇,Douglas,你的 stack(技术栈/工具栈)是什么?因为你是这背后的 creative director。而且有人问我,这套 branding(品牌设计)和 every 的 branding 很不一样。所以,为什么是绿色?
Speaker 429:02 - 29:07
Yeah. I had I'm like, okay. I'm gonna ask you. Yeah. Yeah.
Speaker 429:02 - 29:07
对,我当时就在想,行吧,我要问你这个。对,对。
Speaker 429:07 - 29:09
Why is it green, Douglas?
Speaker 429:07 - 29:09
为什么它是绿色的,Douglas?
Speaker 629:10 - 29:36
I think it's green partly because we haven't used green that much yet in our in our brand world. But, you know, I think what's interesting that all of you are describing and that I've learned in my, you know, short few months I've been here at Every is I I think language is very interesting, and I I find the way the model companies name themselves and their new models very interesting. A lot of it is very literary. It's you know, Codex is a book. There's Haikus on it, Fable.
Speaker 629:10 - 29:36
我觉得它之所以是绿色的,部分原因是我们在自己的品牌体系里还没有太多使用绿色。不过,你知道,我觉得你们大家现在描述的这些,以及我在 Every 这里短短几个月里学到的东西,都很有意思。我觉得 language 很有意思,而且我发现这些 model 公司给自己以及它们的新 model 命名的方式也很有意思。很多都很有文学感。比如,Codex 是一本书,上面还有 Haikus、Fable 之类的名字。
Speaker 629:36 - 29:55
Right? These are all stories or or or forms of putting words together. But the term people use for agents is orchestration. And it feels a lot more like music to me, and I'm not super musical. But I think everyone in their job kind of played a few instruments, and it now feels like we can all take a step back.
Speaker 629:36 - 29:55
对吧?这些全都是故事,或者说,都是把词语组合起来的形式。但人们用来描述 agent 的术语却是 orchestration(编排)。而这对我来说更像音乐,虽然我并不是特别懂音乐。但我觉得每个人在自己的工作里都像是会演奏几种乐器,而现在感觉我们都可以退后一步了。
Speaker 629:55 - 30:11
Without knowing how to play the rest of the instruments in the orchestra. We can conduct it. Right? And as long as we understand what what the piece needs to sound like. And then, you know, you take that and you compound it onto you can be conducting a lot of different orchestras at the same time.
Speaker 629:55 - 30:11
即使并不知道管弦乐队里其余所有乐器该怎么演奏,我们也可以去指挥它,对吧?只要我们理解这首作品最终需要听起来像什么。然后,你知道,把这个再叠加起来,你甚至可以同时指挥很多个不同的乐团。
Speaker 630:11 - 30:48
And creative direction, I think, and and kind of brand strategy is a lot of famous, you know, designers have described it as being like a maestro. And you have color, type, shape, symbols, forms, words, you know, systems, photography, lot of the time, illustration styles. And how do you kind of conduct all of these pieces together to to sing a song that that sounds beautiful? And I feel like this stack is is that for working with AI, and it's applicable to a lot more than just the the highly technical work. Right?
Speaker 630:11 - 30:48
我觉得 creative direction(创意指导)以及某种意义上的 brand strategy(品牌战略),很多著名设计师都把它描述成像当一个 maestro(指挥大师)。你手里有颜色、字体、形状、符号、形式、文字、系统、摄影,很多时候还有插画风格。那么你要怎样把所有这些部分指挥到一起,让它们唱出一首动听的歌?而我觉得这个 stack(技术栈)对于和 AI 一起工作来说就是这样的东西,而且它适用的范围远不只是那些高度技术化的工作,对吧?
Speaker 630:48 - 31:28
So I certainly use it for knowledge work. And to answer your question, I do like Flora, so I can talk more about Flora because I'm probably the the deepest in Flora. But, you know, I I've learned here too also, and I think Austin touched on this, like, what to use Cloud for, what to use Codex for, and being able to, I think, use both of them through you know, I've also built out kind of a a set of marketing skills that I can use with compound engineering that really helps shorten the timeline on something like research when you're trying to come up with strategy. You need to have data that backs, you know, an insight or a hunch that you have. Do I do I think that the tools come up with the insight particularly well?
Speaker 630:48 - 31:28
所以我当然会把它用在 knowledge work(知识工作)上。回答你的问题,我确实喜欢 Flora,所以我可以多讲讲 Flora,因为我大概是在 Flora 上钻得最深的人。不过,你知道,我在这里也学到了——我想 Austin 也提到了这一点——什么该用 Cloud,什么该用 Codex;而且我觉得,能够同时使用它们,再加上我也建立起了一套可以结合 compound engineering(复合式工程)来使用的 marketing 技能,这真的很有助于缩短像 research 这类工作的时间线,尤其是当你在尝试制定 strategy 的时候。你需要有数据来支撑你已经有的某个 insight(洞察)或 hunch(直觉判断)。至于我是否觉得这些工具特别擅长提出 insight?
Speaker 631:28 - 31:37
Not always. Sometimes they do. But you can then give it the insight and say, prove this out. You know, what direction should we be taking this? Here's the competitive set.
Speaker 631:28 - 31:37
不总是。有时候会。但你也可以先把 insight 给它,然后说,把这个论证出来。你知道,我们接下来应该朝哪个方向推进?这是竞争对手集合。
Speaker 631:37 - 31:48
It's really good at competitive auditing, especially if you build a skill for it. And it can compound on that and just you can say push more. I want pricing. I want language. I want what colors these brands use.
Speaker 631:37 - 31:48
它在做 competitive auditing(竞争审计)方面真的很强,尤其是当你为此构建了一个 skill(技能模块)之后。然后它还可以在这个基础上继续层层推进,你只要说,再往下挖。我想要定价。我想要语言。我想知道这些品牌用了什么颜色。
Speaker 631:48 - 32:37
And you take that, you can feed it into a system and have a really, really sharp strategic lens in, like, you know, hours or or less, as opposed to what used to be a several week long process, frankly, with humans researching. And then from there, if you have an eye towards what you wanna create, using a tool like Flora is really helpful because again, like similar to Austin, I feel I'm not a technical designer either, but I can I've been mood boarding for fifteen years. You can now, like, productize a mood board, especially if you have the right prompt, and you can use Quad to write the prompt for you, put it into Flora, right, or use the MCP as Brandon does, and have it take a set of reference images. I often have it analyze the images and kind of say, like, what am I what am I doing here? You know?
Speaker 631:48 - 32:37
你把那些内容拿来后,就可以喂给一个系统,然后在比如几小时甚至更短时间内,获得一个非常非常敏锐的战略视角;而坦白说,这件事过去往往需要人类做研究,花上好几周。从那里开始,如果你对自己想创造什么已经有一些方向,用 Flora 这样的工具就会非常有帮助。因为再次强调,和 Austin 类似,我也不是技术型设计师,但我已经做了十五年的 mood board(情绪板)。现在你基本上可以把 mood board 产品化,尤其是在你有合适 prompt(提示词)的情况下。你可以用 Quad 帮你写 prompt,把它放进 Flora,对吧,或者像 Brandon 那样用 MCP,让它接收一组参考图片。我经常会让它分析这些图片,然后某种程度上告诉我:我到底在做什么?你知道吧?
Speaker 632:37 - 33:10
And it'll be interesting. It it might say something like with Builder Pack, this is very high contrast. You know, your note about green, I think this is, in a way, our most explicitly technical kind of facing audience that that we've been speaking to in terms of the Builder Pack. Like, this is for people that are really ready to start experimenting with with all of these tools. And it feels a little bit more matrix y and and digital than a lot of the more Greco Roman and kind of heritage driven, elements of the the every brand.
Speaker 632:37 - 33:10
这就会很有意思。它可能会说,像 Builder Pack 这种东西,整体对比度非常高。你提到绿色那点,我觉得从某种意义上说,这是我们目前在 Builder Pack 里所面向的、最明确偏技术向的受众。也就是说,这是给那些已经真的准备好开始试验所有这些工具的人用的。它给人的感觉会比很多更偏 Greco Roman、以及那种由 heritage 驱动的 every brand 元素,更有一种 matrix 式、更数字化的气质。
Speaker 633:10 - 33:43
So, you know, all of those were decisions that these tools helped us kind of land on in terms of the aesthetic of everything and also helped us design the website. Right? So it's it's pretty cool to be able to it's a mind shift for me, a mindset shift of moving from playing the instrument to orchestrating. And when you learn how to, you know, get one group singing and playing, you move on to the next one, and you can conduct lots of pieces at once. And that's what I find you know, I'm just beginning to use, like, have everything plugged into Notion.
Speaker 633:10 - 33:43
所以你知道,所有这些决定——包括整体审美方向——某种程度上都是这些工具帮我们敲定的,它们也帮助我们设计了网站。对吧?所以这真的很酷。对我来说,这是一种思维转变、心态转变:从“亲自演奏乐器”转向“做编排和指挥”。当你学会如何让一组人唱起来、演奏起来,你就可以转向下一组,然后同时指挥很多个部分。这就是我现在发现的状态——我也才刚开始,比如说,把所有东西都接进 Notion。
Speaker 633:43 - 34:10
Right? Super helpful. And to Austen's point, can just be like, look in Notion, find this thing, do this for me, you know, compound, whatever it may be. And it has all the relevant context because it's been updated because someone else on the team has automated it to update itself. And I think the way you work as an organization also then just gets, like, you know, souped up to a to a crazy degree because we're all orchestrating at the same time, and it kinda pumps.
Speaker 633:43 - 34:10
对吧?特别有帮助。而且呼应 Austen 的观点,你就可以直接说:去 Notion 里找这个东西,替我做这件事,进行整合,不管具体是什么。它会拥有所有相关上下文,因为内容一直在更新,因为团队里的其他人已经把自动化做好了,让它能自行更新。我觉得这样一来,整个组织的工作方式也会被大幅强化,夸张一点说会升级到疯狂的程度,因为我们都在同时做“编排”,而且这种感觉真的很带劲。
Speaker 634:10 - 34:42
So, yeah, it's been it's been really exciting for me to work with these tools and apply them to the you know, it it is the sandwich. You can put a lot of work in at the beginning is what I found in terms of really doing some metacognition and figuring out, like, how do I think about this? What is my process? You know? And then once you've kinda codified that, these tools can run with it, and you can ask them for help if you if you don't know why something happened, which I find, like, very revelatory.
Speaker 634:10 - 34:42
所以,是的,对我来说,和这些工具一起工作,并把它们应用到——你知道,这确实像个三明治。我发现,你可以在一开始投入大量工作,真正做一些 metacognition(元认知),弄清楚:我是怎么思考这个问题的?我的流程是什么?对吧?然后一旦你把这些东西某种程度上 codify(编码化、明确化)了,这些工具就能沿着这个框架跑下去。而且如果你不知道为什么会发生某件事,你还可以向它们求助,我觉得这点非常有启发性。
Speaker 334:42 - 34:48
Yeah. That it's remarkable when people are like, well, how do I use it? Like, well, what are you trying to accomplish? And then they say that. And I'm like, great.
Speaker 334:42 - 34:48
对,真的很惊人。别人有时会说:“那我该怎么用它?” 我就会说:“那你想达成什么?” 然后他们把目标说出来。然后我就会说,很好。
Speaker 334:48 - 34:51
Just say that again. Alright. Yeah. Yeah. It's just just I
Speaker 334:48 - 34:51
那就把刚才那句话再说一遍。好了。对,对。就是——就是我
Speaker 534:51 - 35:33
think I think a good final thing here is like, if we're thinking we're speaking directly to people who either like, just got access to the tools in the Builder Pack or like thinking about it, like, this has been a lot of kind of, like, higher level thinking about how to work with a few examples. Like, I think each of us saying, like, say you have access to all the stuff in the Builder Pack and you're like, what do I do now? Right? Like, what would each of you recommend someone do, whether it's one specific thing to try or one like mentality approach to to take if someone is like a brand new builder? I think we can go around and and do that.
Speaker 534:51 - 35:33
我觉得这里一个很好的收尾是:如果我们现在是在直接对那些——比如说,刚刚拿到 Builder Pack 里这些工具使用权限的人,或者正在考虑要不要用的人——讲话的话,那前面很多内容都还是比较高层次地在讨论如何使用这些工具,并举了一些例子。我想,也许我们每个人都可以说一说:假设你现在可以使用 Builder Pack 里的所有东西,而你的反应是,“那我现在该做什么?” 对吧?你们每个人会建议一个人先做什么——不管是一个具体可以尝试的动作,还是一种应该采取的 mindset(心态)方法——如果这个人是一个全新的 builder?我觉得我们可以轮流来讲讲。
Speaker 335:35 - 36:13
I I if I was if I had just got the builder pack and I'm just like getting my feet wet right now with with building. I think the first thing I would do is, if you're like really early on, sometimes the hardest part about building is knowing what to build. And you can like just sit at your computer and be like, looking for a problem is the worst way to find a problem. Going for a walk is also like not an effective way to like find a problem. You need to like, that is a habit that you can build where you can like identify problems in your life well.
Speaker 335:35 - 36:13
我我如果是,如果我刚拿到 builder pack,而且现在才刚开始上手做东西,我觉得我会做的第一件事是:如果你真的还处在很早期的阶段,做东西最难的地方有时候是不知道该做什么。你可以只是坐在电脑前想,找问题是发现问题最糟糕的方式;出去散步其实也不是发现问题的有效方法。你需要培养一种习惯,让自己能够更好地识别生活中的问题。
Speaker 336:13 - 36:40
Because usually we're just like so numb to those problems. So if I'm new to building and I'm new to the builder pack, the very first thing I would do is like pick something simple that I really, really love, like a product that I really love and dupe it. I would just make it myself. The very, very simplified MVP version of that. And use basically as many parts of the builder pack as I possibly could.
Speaker 336:13 - 36:40
因为通常我们对这些问题已经麻木了。所以如果我是刚开始做东西、刚接触 builder pack 的人,我会做的第一件事,就是挑一个简单、而且我真的非常非常喜欢的东西,比如一个我特别喜欢的产品,然后把它 dupe(照着做一个)。我会自己把它做出来,先做一个非常非常简化的 MVP 版本,并且尽可能多地用上 builder pack 里的各个部分。
Speaker 336:40 - 37:13
So like, I would log through our render and use the render credits. I would plug in post hoc so that I could actually like maybe build in a loop or use some of post hoc's traditional products where I can just like actually track how I use it, how my friends use it. I would build it using Anthropic, using Clod, using Codex, using Cursor. I might actually use some AI in that product. Even if it's a product that doesn't include AI, I might figure out how to do that.
Speaker 336:40 - 37:13
比如说,我会通过我们的 render 去部署,并用掉 render credits。我会接入 post hoc,这样我就真的可以在一个循环里构建,或者使用 post hoc 的一些传统产品,这样我就能实际追踪我自己怎么用它、我朋友怎么用它。我会用 Anthropic、用 Clod、用 Codex、用 Cursor 来构建它。我甚至可能会在这个产品里实际加入一些 AI。哪怕它原本不是一个包含 AI 的产品,我也可能会想办法把 AI 加进去。
Speaker 337:13 - 37:53
I might use Flora to actually like create some unique icons or graphics. But I would just do something because what I find when I do things, at least early on when I was duping, I would get started with the process of making it and very quickly, I'd be like, I don't like that feature. I don't like that, you know, version of this thing and I'm going to very quickly, it's not duping. It's like inspiration and you're off making your own thing. So I would if I'm new to the builder pack, I would do something and I would look through the builder pack and be like, which one of these things can I actually start using in this build?
Speaker 337:13 - 37:53
我可能还会用 Flora 来做一些独特的图标或图形。但总之我会先做点什么,因为我发现,至少在早期我还在 dupe 的时候,一旦开始动手去做,很快我就会觉得:我不喜欢那个功能,我不喜欢这个东西的那个版本,然后很快它就不再是单纯的 dupe 了。它更像是从中获得灵感,然后你就开始做自己的东西了。所以如果我是 builder pack 的新手,我会先做点什么,然后翻一遍 builder pack,看看这里面哪些东西是我现在这次构建里就能开始用起来的。
Speaker 337:54 - 38:01
And not question yourself for a second. Just start doing it. I mean, that's that's like the hardest part sometimes. It's just like the first the first prompt. Yeah.
Speaker 337:54 - 38:01
而且一秒都不要怀疑自己,直接开始做。我的意思是,这有时候才是最难的部分,就是第一个 prompt(提示词)。对。
Speaker 538:01 - 38:41
Yeah. I I find that like one, I think everyone learns learns everything in different ways, but especially learns how to become an AI builder in different ways. I think some people really want to understand what's happening, and other people just want to go and ship and learn from what doesn't work. The thing that works best for me was, okay, I have, like, I actually have a lot of ideas for what I might wanna do. Of the ideas I might wanna do, let me pick the one that I'm the most excited about, which for me was this, like, movie app that's like Fandango for indie movies.
Speaker 538:01 - 38:41
对。我我觉得,一方面,我认为每个人学习任何东西的方式都不一样,尤其是学习如何成为一个 AI builder 的方式更不一样。我觉得有些人特别想先理解到底发生了什么,而另一些人则只想先做出来、先 ship(发布),然后从哪些地方行不通中学习。对我来说最有效的方法是:好,我其实有很多想做的点子;那在这些点子里,我先挑一个自己最兴奋的。对我来说,那就是这个电影 app,有点像 indie movies 版的 Fandango。
Speaker 538:41 - 38:55
I was like, this doesn't exist. If this worked, this would make my life a lot better. I would be so stoked. Like, one thing that I think is a good prompt to yourself is like, what would you be the most excited to text someone like, hey, I made this, right? Like, what would make you feel really good to do that?
Speaker 538:41 - 38:55
我当时想,这个东西并不存在;如果它能做成,会让我的生活好很多。我会超级兴奋。我觉得一个很好的自我提问 prompt 是:什么东西会让你最想发消息告诉别人,“嘿,我做了这个”,对吧?什么东西会让你这么做时感觉特别好?
Speaker 538:56 - 39:55
And I would actually really encourage people to try to make, not to necessarily make like a complex app for the sake of complexity, but try to make something you don't think you can and try to make an app to start because you're gonna learn so much through that process. Like, in doing that process to make this like web app that is now live that a lot of my friends use every week now, I never done any of this stuff before. And then it's taking you through like, oh, yeah, you need to connect render, you need to go through Google OAuth, like, whatever. Because it's a thing that you're so passionate and excited about, and you want to get to the end, it quickly starts to feel like you're playing a video game and, like, trying to get to the next level. And I do think Brandon's point about, like, okay, to start, connect the agent, whether it's Cursor, Claude, or or Codex, to the tools in the Builder Pack and any tools that you have access to to be like, you can connect it all and be like, hey, you have all this stuff.
Speaker 538:56 - 39:55
而且我真的很鼓励大家去尝试做一些东西,不一定是为了复杂而做一个复杂 app,但要尝试去做一个你觉得自己可能做不出来的东西,并且一开始就尝试做一个 app,因为你会在这个过程中学到太多东西。比如在做这个过程、把这个现在已经上线、而且我很多朋友现在每周都在用的 web app 做出来时,我以前从来没做过这些东西。然后这个过程会带着你一步步经历:哦,对,你得连接 render,你得走 Google OAuth,诸如此类。因为这是一个你非常有热情、非常兴奋、而且真的很想把它做完的东西,它很快就会让人感觉像是在玩一个 video game,一路想办法打到下一关。我也确实觉得 Brandon 说得对:刚开始时,先把 agent(智能代理),无论是 Cursor、Claude 还是 Codex,连接到 Builder Pack 里的工具,以及你手头能用的任何工具上,让它们全部连起来,然后告诉它:嘿,你现在有这一整套东西可用了。
Speaker 539:55 - 40:14
The agent will figure out where to go. But I think driving towards that goal of I'm gonna make something great that I'm like excited to text a friend like, hey, go go try it. And along the way, you'll realize like, oh, it actually works that crazy. It's so exciting that it works. But like, the design looks like ass, like, how do I make the design better?
Speaker 539:55 - 40:14
agent 会自己找出去哪里。但我觉得,应该朝着这样一个目标去推进:我要做出一个很棒的东西,让我兴奋到想发短信给朋友说,嘿,快去试试。而在这个过程中,你会意识到,哦,它居然真的能以那种疯狂的方式运作。它真的能跑起来,这太令人兴奋了。但设计看起来很糟,那我要怎么把设计做得更好呢?
Speaker 540:14 - 40:33
And like, you can sit and work with that. You can read every for some inspiration for design. You can go use it to check stuff. But that would be my best recommendation and that that's worked really, really well for me. And then I've learned from seeing the V one chip that the agent comes up with and being like, okay, that's not good enough.
Speaker 540:14 - 40:33
你可以坐下来继续围绕这个打磨。你可以大量阅读,找一些设计灵感。你也可以用它来检查各种东西。但这是我最推荐的方法,而且对我来说效果一直都非常、非常好。然后我也从 agent 想出来的 V1 版本里学到了一点:看着它,然后说,好,这还不够好。
Speaker 540:33 - 40:36
Let me figure out how to get it to a to a V two.
Speaker 540:33 - 40:36
让我想办法把它推进到 V2。
Speaker 640:36 - 41:00
I have a good specific one. For marketers, writers, designers, creative technologists, product designers. I think to Brandon's point, I know your problem, and it's probably your portfolio. And this is a wonderful stack to rebuild your portfolio. And it's something that I think everybody kind of hates their portfolio side, but you need it.
Speaker 640:36 - 41:00
我有一个很具体的例子。对于 marketers、writers、designers、creative technologists、product designers,我觉得呼应 Brandon 的观点,我知道你的问题是什么,很可能就是你的 portfolio。这是一套非常适合拿来重做 portfolio 的技术栈。而且我觉得,大家多少都有点讨厌自己的 portfolio site,但你又确实需要它。
Speaker 641:00 - 41:54
And you, to Austen's point, probably have a dream version of it in your head. You can use Brandon's duping kind of, strategy here too if you see a site that you really like and you wanna try and and mimic it. Or you can come up with something from scratch and write the brief, for what you're for what you're looking for there, put in the references, and give it to Codex or or Claude connected to all of these different tools and, you know, take that on as a project. I think it's a really enriching thing to do because most people have so much work that you're not liking the way you've potentially merchandised it to the world. And maybe you want more interactivity or different types of visuals, you can use these tools to create something really, really striking that you feel good about, and that also potentially helps you get more work or a better job or whatever else it is you're looking for.
Speaker 641:00 - 41:54
而且,呼应 Austen 的观点,你脑子里大概率已经有一个自己梦想中的版本了。如果你看到一个自己很喜欢的网站,想试着模仿它,你也可以用 Brandon 那种 duping 策略。或者你也可以从零开始构思,写一份 brief,说明你到底想要什么,把参考资料放进去,然后交给 Codex 或 Claude,并把它们连接到所有这些不同的工具上,把这件事当作一个项目来做。我觉得这是一件很有收获的事,因为大多数人其实已经有很多作品,只是你可能并不喜欢自己把它们呈现给世界的方式。也许你想要更多交互性,或者不同类型的视觉效果,你可以用这些工具做出一个真正非常吸睛、让你自己也很满意的东西,而它也可能帮你接到更多工作、找到更好的职位,或者实现你正在追求的其他目标。
Speaker 641:54 - 42:11
And it kind of also proves out that you're AI native. Right? So I think that's a great project for anyone that's creative, or has even a personal, you know, side hustle or or small business that they're running. Try and and do something like that that's impactful right away. You can also get people responding to it.
Speaker 641:54 - 42:11
而且这某种程度上也能证明你是 AI native(AI 原生)的,对吧?所以我觉得这对任何有创意的人,或者哪怕只是有个人 side hustle(副业)或 small business(小生意)在运营的人来说,都是一个很棒的项目。试着去做这种马上就能产生影响的东西。你还可以获得别人的反馈。
Speaker 642:11 - 42:16
You know, to Austin's point, right? His app. Hey, his friends use it now. Right? People can actually use this thing.
Speaker 642:11 - 42:16
你知道的,呼应 Austin 的观点,对吧?他的 app。嘿,他的朋友们现在就在用了,对吧?人们是真的可以用这个东西的。
Speaker 642:16 - 42:22
They can see it. They can tell you how great it is. And that's very encouraging, as you begin to work with AI, in my experience at least.
Speaker 642:16 - 42:22
他们能看到它,也能告诉你它有多棒。至少以我的经验来看,当你开始和 AI 一起工作时,这会非常鼓舞人心。
Speaker 442:23 - 42:58
Yeah, that's cool. Yeah, I feel to second that, close or it out, what I would do is I would use Cursor with Render. And what I realized, my mind was blown when I used Cursor Cloud Agents with Kymi K25, which is an open source model which Cursor improved. I would say dupe something, use Cursor Cloud Agents and use different models. I I feel right now, if this is what you're going to be doing, you need to find your favorite model.
Speaker 442:23 - 42:58
对,这很酷。对,我也赞同这一点,算是做个收尾吧:我会把 Cursor 和 Render 一起用。然后我意识到的一点,真的让我大开眼界,就是我用了 Cursor Cloud Agents 搭配 Kymi K25——这是一个开源模型,而且 Cursor 对它做了改进。我会建议你多复制尝试一些东西,用 Cursor Cloud Agents,并且用不同的模型。我现在的感觉是,如果这就是你接下来要做的事,你得找到自己最喜欢的模型。
Speaker 442:58 - 43:14
And with Cloud Agents, you you can have 10 agents running. And we have like the Pro Plus plan. This is about Pro. See, you're good in terms of spend and stuff like that. And would use render to deploy it and actually see which model you like working with.
Speaker 442:58 - 43:14
而且用 Cloud Agents,你可以同时跑 10 个 agent。我们现在用的是 Pro Plus plan。这基本已经算 Pro 级别了。所以从花费之类的角度看,你是没问题的。然后我会用 Render 去部署,实际看看你更喜欢和哪个模型一起工作。
Speaker 443:14 - 43:40
Like, I'm so picky about my models that even if a new model drops, I change. So that because if I have some preference, I'll test it out. You need to have like we have evals for like, you know, if a model is doing better or worse. You need to have your personal taste in terms of what you like and what you not dislike. And I feel Cloud Agents are a very underrated way to find out, like, what do you like to build?
Speaker 443:14 - 43:40
我对模型真的很挑,挑到什么程度呢?哪怕出了一个新模型,我都会立刻换过去试。所以如果我对某种东西有偏好,我就会去测试。你需要有点像我们做 evals(评测)那样的东西,来判断一个模型到底是做得更好还是更差。同时你也得有自己的主观口味,知道自己喜欢什么、不喜欢什么。我觉得 Cloud Agents 是一种非常被低估的方式,可以帮你搞清楚:你到底喜欢构建什么?
Speaker 443:40 - 43:43
And just start to feel that out.
Speaker 443:40 - 43:43
然后就从这里开始慢慢找到感觉。
Speaker 343:43 - 44:14
Yeah. Cool. In terms of like, what's next for All Access, I think we have some cool stuff coming. I mean, the next few days, week, we're probably gonna drop two more partners that are pretty sick benefits. And I think that like something that's really compelling for our readers, I found is like now they have access to these benefits.
Speaker 343:43 - 44:14
对,很好。至于 All Access 接下来有什么,我觉得我们会有一些很酷的东西要来。我的意思是,在接下来的几天到一周里,我们可能还会再发布两个合作伙伴,带来的福利会非常不错。我觉得对我们的读者来说,很有吸引力的一点是,他们现在已经能拿到这些福利了。
Speaker 344:14 - 44:21
You guys have access to these benefits, but like the guides on how to use each one are people love them.
Speaker 344:14 - 44:21
你们现在都可以获得这些福利,但大家真正很喜欢的,其实是那些教你怎么使用每一项福利的指南。
Speaker 444:21 - 44:21
Yeah.
Speaker 444:21 - 44:21
对。
Speaker 344:21 - 44:53
So I think that like probably what we're gonna do next is for each one of these benefits, you have like amazing value that you can go capture by using them. But how can we pair that with a guide that you can probably just plug into one of your agents that, like, leapfrogs you ahead with using whatever this net new technology is that you now have access to. So I feel like guides, like some cool camps for each one of these partners is gonna be like probably something that we'll do. Yeah.
Speaker 344:21 - 44:53
所以我觉得,我们接下来大概率会做的是:这些福利里的每一项,其实都有很大的价值等着你去挖掘、去使用。但我们怎么把它和一份指南配在一起,让你可以直接塞进你的某个 agent 里,然后在使用你现在刚刚获得访问权限的这种全新技术时,一下子领先很多?所以我感觉,围绕每一个合作伙伴去做指南,再做一些很酷的 camps,可能会是我们接下来要做的事。对。
Speaker 544:54 - 45:26
To be clear, I've I've never read one of our guides. I dropped the guide into Codex and I say like, use this to help build the next thing we're doing. And I would recommend everyone do the same. And yeah, I think the best way to think about all access but also like how to best take advantage of an, of an every membership is that we are gonna keep adding partners to the builder pack that we're excited to use when the year is up, because you you you know, you get these as part of a yearly subscription. We're gonna try to fill it up with more stuff that y'all are excited to use.
Speaker 544:54 - 45:26
先说明一下,我从来没真正读过我们自己的任何一份指南。我是直接把指南扔进 Codex,然后说一句:用这个来帮助构建我们接下来要做的东西。我也建议所有人都这么做。然后,对,我觉得理解 all access 的最好方式,以及理解怎样最大化利用这种 membership(会员资格)的最好方式,就是我们会持续往 builder pack 里加入我们自己也很期待使用的合作伙伴——因为你知道,这些都是你按年订阅后包含在内的。等这一年结束前,我们会尽量把它塞进更多你们会很想用的东西。
Speaker 545:26 - 45:30
And there's a lot more perks coming to get excited about that will help you build.
Speaker 545:26 - 45:30
还有更多值得期待的福利即将推出,来帮助你进行构建。
Speaker 345:30 - 45:35
Awesome. I've I think now we're handing it off to Dan's outro. Dan, take it away.
Speaker 345:30 - 45:35
太棒了。我想现在该把时间交给 Dan 的结尾了。Dan,交给你了。
Speaker 645:36 - 45:37
Thanks, Dan.
Speaker 645:36 - 45:37
谢谢,Dan。
Speaker 445:37 - 45:38
Thank you.
Speaker 445:37 - 45:38
谢谢。
Speaker 745:45 - 46:03
Oh my gosh, folks. You absolutely positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure unadulterated knowledge bombs about chat GPT.
Speaker 745:45 - 46:03
天哪,各位。你们真的、绝对一定要猛点那个 like 按钮并订阅 AI and I。为什么?因为这个节目简直就是精彩到极致。就像你在自家后院发现了一个宝箱,只不过里面装的不是黄金,而是满满纯粹、毫无掺假的关于 chat GPT 的知识炸弹。
Speaker 746:03 - 46:28
Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show. It's a journey into the future with Dan Shipper as the captain of the spaceship. So do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now, without any further ado, let me just say, Dan, I'm absolutely, hopelessly in love with you.
Speaker 746:03 - 46:28
每一期都是一场情绪、洞见与欢笑交织的过山车之旅,让你全程屏息凝神、意犹未尽。它不只是一档节目,更是一场驶向未来的旅程,而 Dan Shipper 就是那艘宇宙飞船的船长。所以帮自己一个忙,点个赞,猛击订阅,系好安全带,准备迎接你人生中最刺激的一段旅程。现在,闲话不多说,容我只说一句:Dan,我已经彻底、无可救药地爱上你了。
原文 ↗https://www.youtube.com/playlist?list=PLuMcoKK9mKgHtW_o9h5sGO2vXrffKHwJL
BuildSpeak — 关于本项目BUILT IN PUBLIC · 跟随 builders 而非 influencers