I stole the AI product stack of the top 1% product managers for you (full tutorial)

1. 0:00 – 1:30

   Intro

   1. RWRachel Wolen0:00

      I run my whole day out of Claude Code and Cursor. These are my AI superpowers, and I'm encouraging my entire team to use them.

   2. AGAakash Gupta0:07

      This is Rachel Wolen, the Chief Product Officer at Webflow, the $4 billion web giant used by companies like TED Talks, SoundCloud, Meetup.

   3. RWRachel Wolen0:17

      So it's almost like having a data scientist in my pocket.

   4. AGAakash Gupta0:19

      I think this is the art of building amazing AI-native product.

   5. RWRachel Wolen0:23

      Gets rid of a lot of the junk in my inbox first, and then it will actually create drafts for a few people that it thinks I need to actually send emails to. And I think that's the one thing I would say about a lot of, like, building an agent is trying it out and then going and adjusting what you want the agent to do.

   6. AGAakash Gupta0:38

      This is the roadmap to becoming a great product leader. There are tons of tutorials about Claude Code and Cursor for ICPMs, but what about product leaders? Today's episode is a masterclass. What is this concept, and how can CPOs effectively do IC work?

   7. RWRachel Wolen0:54

      To me, IC CPO means as a leader, you are able to get your own answers to practically any question.

   8. AGAakash Gupta1:01

      We've just showed all these amazing workflows. How do you set up your organization to work this way?

   9. RWRachel Wolen1:07

      I think one of the things I would first assume is that-

   10. AGAakash Gupta1:13

       Really quickly, I think a crazy stat is that more than 50% of you listening are not subscribed. If you can subscribe on YouTube, follow on Apple or Spotify podcasts, my commitment to you is that we'll continue to make this content better and better. And now on to today's episode.

2. 1:30 – 3:25

   What Is IC CPO?

   1. AGAakash Gupta1:31

      Welcome to the podcast.

   2. RWRachel Wolen1:32

      Thanks, Aakash. Great to be here.

   3. AGAakash Gupta1:33

      What are we gonna do today?

   4. RWRachel Wolen1:34

      We are gonna go through some of my workflows, my agentic chief of staff, um, and a bunch of different ways that I use Claude Code and Cursor, and then I'm going to walk you through what it's like to actually build AI-native products in the wild, and we're in the middle of getting a product, a new code gen product, ready to go. And so I will walk you through the good, the bad, and the ugly of trying to get a new AI-native product out into the wild and into our customers' hands.

   5. AGAakash Gupta2:02

      Awesome. So as you alluded to there, there's really two sides to AI product leadership. There's being a productive AI product leader, and there's shipping AI-native features. So in the productivity bucket, let's start here at this concept of IC CPO. What is this concept, and how can CPOs effectively do IC work?

   6. RWRachel Wolen2:21

      Yeah, so to me, IC CPO means that as a leader, you are able to get your own answers to practically any question. And if you stated that as a goal, then you kind of have to work backwards and look at, how do I make sure my data is in shape where I, as well as anybody else on my team, could go and self-serve answers? And that is a large task to undertake, I can tell you that from experience, and we're still in, in the middle of that. Um, I think the second is making sure your team has the right tools for what they're trying to accomplish and then can even stair-step their way up. And then I think the third is really, um, you know, figuring out how and when to model for your team, not because you expect them to copy your workflows, but you want them to be inspired. I think that part of, you know, being a great leader today is also being a great IC and getting your hands as dirty as you can, um, carving out time to experiment and showing your team that it's okay to experiment and for sometimes it works, sometimes it doesn't, but that's part of, you know, like, building today.

   7. AGAakash Gupta3:24

      Amazing.

3. 3:25 – 7:03

   Building Agentic Chief of Staff

   1. AGAakash Gupta3:25

      So let's start with Claude and Cursor. Can you walk us through how you're doing some of this IC CPO activity through them?

   2. RWRachel Wolen3:31

      Yeah, absolutely. So what I've done is I've built out what I call my agentic Chief of Staff, and this is a combination of a set of Claude Code agents rather, as well as an app that I'm building that I use on a day-to-day basis. So first, I'll kind of walk you through how I use Claude Code, um, and this is, by the way, in Cursor. You could do this in any IDE. I also, depending on what my task is, sometimes I will use Cursor and the Cursor agent. A lot of times if I'm, like, trying to start something, a project from scratch, I will install from Cursor. I will also sometimes use, uh, Codex, especially if it's, like, a complex, um, type of task where I'm trying to understand content from our monorepo. And so, you know, I, I basically am running out of terminal for a bunch of different tasks. And so what I've done is I've created a set of agents, um, and I'm, I'm kind of, like, constantly adding to those agents. So I'll just show you what, like, an agent looks like, and this was generated by Claude, and we'll, we'll go through, like, how to actually generate an agent. But this i- this one is, like, understanding the priority of a calendar event-

   3. AGAakash Gupta4:40

      Mm-hmm

   4. RWRachel Wolen4:41

      ... and trying to decide if it is truly important. Um, and so what I'll show you is, like, I, I actually was running this earlier today and, you know, you're trying to decide, is this part of my priorities, and then trying to filter out noise and then looking for different types of meetings that are very important. Um, and then also really, y- like, this was all generated by Claude, by the way. Um, and then what I'm doing, uh, so I've ran this earlier today. Let's see. Hopefully it doesn't have anything too crazy in here. And what I [laughs] you know, w- I, I think I looked at this before I, I got in here. And so what I did was first, like, I asked it, uh, earlier today, I, I ran it 'cause it takes a little bit of time. I, I asked to look the last two weeks, and I said, "Can you analyze my calendar, uh, like, how I, how I spent my time?" Um, so, "Analyze my calendar for the week. How did I spend my time? Where could I have been more effective at delegating?" Right? So this is something that I do usually, like, once a week, but I also will run this once a day as well. Um, and then fir- first it kind of gave me, like, "Here are delegation opportunities." These are actually right. I ended up not attending this meeting because I needed to get ready for this podcast.

   5. AGAakash Gupta5:52

      [laughs]

   6. RWRachel Wolen5:53

      Um, [laughs] funny enough. Um, this is a meeting that usually I do send one of my directs to that is our, our growth lead. Like, this is spot on, right?Um, this is a demo lab that we run, um, called Alpha Arcade, and I usually don't end up attending that one. Um, merge council is something that I have somebody on my direct team. So this, this is all, like, completely correct. And then it also identified, like, RAG flags, like, where I'm, like, double and triple booked. Um, you know, it, like, said, "Hey, you're not, like... You're context switching too much." I mean, this is, this is correct, right? And this is something that I give to my EA, and I'm telling her, "Hey, like, this is, this is kind of what we're seeing, and we'll, we'll start with this at the beginning of the week." Um, and then, you know, it also-- I also like to look forward at the following week and be like, "Hey, what can we do to improve things?" And, you know, it, it... a lot of it is like, "What, what do you recommend I should cut for next week?" Um, I'm not sure every- I agree with everything in there, but a lot of this is, like, a first pass, and it's organizing it in a way that, you know, makes a ton of sense to me. And so, and, and this is just from basically running the agent and giving it that one line.

4. 7:03 – 9:30

   Calendar Agent Demo

   1. AGAakash Gupta7:03

      That's cool. So you've got one of your Chief of Staff agents. What else are you building around the Chief of Staff space?

   2. RWRachel Wolen7:09

      Yeah. So I do the same thing in email, so it has complete access to my email triage, and, and this is spot on. So basically-

   3. AGAakash Gupta7:16

      Mm-hmm

   4. RWRachel Wolen7:16

      ... I, I asked it, um, to triage my email, and I, I think I ran it a couple of times accidentally. What it does is it gets rid of a lot of the junk in my inbox first, but it will first go through and run it. It runs a triage first, and then I tell it what to archive. So I don't want, like, calendar notifications or marketing or systems messages. Then it, like, pins the messages that are kept, and then it will actually create drafts for a few people that it thinks I need to actually send emails to, right? I don't want it sending emails on my behalf. That's not the point. But I do think that there are, like, opportunities to... where it's, like, an email that's been sitting in my inbox and sees me. It kind of is actually, like, watching the behavior in my inbox. Um, and then, you know, ultimately I'll get into a, a much, much healthier state. The other thing that was, like, funny, I ran it this morning, and I had a, a meeting with someone [chuckles] that it didn't have a meeting link, and it called it out. So it, it-

   5. AGAakash Gupta8:07

      Yeah

   6. RWRachel Wolen8:07

      ... it's, like, kind of little things like that where maybe there are mistakes. Um, and it's not typically acting on my behalf. It's just running the triage, um, for me. So it, it recommends it would archive 40 emails, right? It would keep it in the inbox. And then I basically say yes or no to go and, and run those actions.

   7. AGAakash Gupta8:25

      This is epic. So how would somebody set this up? How do they connect their email and calendar to Claude Code and give it access?

   8. RWRachel Wolen8:31

      Yeah. So I basically set up a token in Google. Um, I'm not gonna show you my exact env, uh, env setup, but I, I store-- I basically generated a token on Google Cloud, and then I store that in my .env file, um, which is... I'm not gonna show you the actual file, because it has all of my tokens-

   9. AGAakash Gupta8:52

      [laughs]

   10. RWRachel Wolen8:52

       ... and I can regenerate them. Um, but basically I have, like, an env file, and then that i- the, the reason why it has a, uh... This is basically ignored by Cursor, right? This is also ignored by Git. So I have this as, like, a GitHub repo that I maintain, but it doesn't... This, this particular file does not get synced, and so there are variables that are in that file, and, uh, and Claude Code, like, generates it for you. So it's not... Uh, I basically tell it, "Hey, generate a variable, uh, for Gmail," right? And it'll say, "Okay," and it's in this file. And then it will go... Then I go, and I generate it on, uh, Google in, in

5. 9:30 – 13:01

   Email Triage Agent

   1. RWRachel Wolen9:30

      the console.

   2. AGAakash Gupta9:30

      Awesome. So we got the high level overview. You're getting your full Chief of Staff agents. You also have an analytics agent. Can you show us that?

   3. RWRachel Wolen9:38

      I do, and this one's really fun. Um, so I figured I would show one thing that is more fun. So I, my wife has a, a company called Shirts. It is an AI, uh, T-shirt design company called Shirts.com. This is, uh, my wife's company. It's called Shirts.com. It is an AI T-shirt generator. Uh, you know, I generated a T-shirt. We're talking about answer engines. This is, you know, a fun T-shirt [chuckles] I generated for my, uh, for my team, 'cause we just launched an answer engine, uh, optimization product. And what I'll show you in... C- I assume you can see my cursor here.

   4. AGAakash Gupta10:15

      Yep.

   5. RWRachel Wolen10:15

      Great. The way that I run this is I can actually query Snowflake out of Claude. This is obviously a workspace where we have our website running out of Webflow. Then we've got a number of different sites that you can see, and I can ask it questions about those sites and when was the last time that... You know, how many HTML blocks? Like, what, uh, when was the last time it was published? What features is it using? And this is really useful for me. You can imagine if I, like, go into a customer meeting, and I want to know what they're using on Webflow. That is sometimes, like, not something I want to go and bother a data scientist with.

   6. AGAakash Gupta10:48

      Yeah.

   7. RWRachel Wolen10:49

      They have lots more important problems than this to go and, and tackle. But it is useful, and it's something that I want to enable anybody on the team to do. So one of the things that I was talking about, um, previously was that I think a big... Y- like, my vision for the, our, our insights team was to be able to self-serve any insight, um, that is kind of like a, an insight where maybe it has a yes or no answer or it has, like, a very specific piece of data that you're trying to collect, right? Like, in this case, I'm trying to collect information about the websites in this particular workspace for Shirts, right? And so, like I said, Shirts is a AI design generator, but I'm like, "Oh, this is interesting that we're not actually using very much..." Uh, you know, like we... It's not a very complex site yet. It's just a very simple vanilla JavaScript Webflow implementation. This is then, uh, informed by this directory. So we've actually gone, and we've done a bunch of analysis of all of our models, and then we've started to document our models. You kind of have to do this in order to be able to get good outputs from your Snowflake when you're sending natural language queries. The other thing that I've done here is I've basically set up MCP servers for Snowflake, uh, for Tableau. Um, Snowflake and Tableau, I believe, are not officially supported, uh, repos. And I basically, the way that I set it up was I, I just fed the repo to Claude Code, and I'll, I'll, I can... We can put those in the show notes. And then I said, "Hey, I want you to use this MCP server." And then all it does is it authenticates with your credentials.

   8. AGAakash Gupta12:20

      Mm-hmm.

   9. RWRachel Wolen12:20

      So it uses my SSOCredentials. And so I'm not, like, sharing any data that I don't already have access to.

   10. AGAakash Gupta12:28

       Mm-hmm.

   11. RWRachel Wolen12:28

       And this is all being done locally, and it's being run through a work, uh, you know, basically through, through our, our work, um, Anthropic account. So, you know, I think that's, like, a big thing to really be thinking about. What do you have access to? Are you trying to give too much information to the, you know, to the model? And so a lot of this, I think, is also, like, a good exercise in understanding privacy, um, and really trying to think about, uh, you know, how... Like, when you're building software even, what, what information do you want to give to the model-

   12. AGAakash Gupta13:00

       Mm

   13. RWRachel Wolen13:00

       ... and are you comfortable with?

6. 13:01 – 13:48

   Linear Ad Start

   1. AGAakash Gupta13:01

      Yeah. You see that right there? That's Linear, today's sponsor. If you haven't heard of Linear yet, check it out. It is the task management platform that is dethroning Jira. Teams like OpenAI, Vercel, Perplexity, they all use Linear. Why? I had Nan Yu, the head of product, on the podcast, and he explained that those teams value speed. Linear is built for speed. If you have engineers who wanna be operating at that next level of speed, you need Linear. And even if you're a product manager, you need Linear, because Linear has Linear Agents, where you can call Devin, call Claude Code, call Cursor to complete a task for you. Imagine you have a ticket. You ask an agent to do it. The on-call engineer reviews it, and it's live in production. PMs are getting closer to the bare metal of the product with Linear, today's sponsor, so check it out at linear.com/aakash.

7. 13:48 – 17:27

   Analytics Agent (Snowflake + MCP)

   1. AGAakash Gupta13:49

      So how would you use the analytics agent?

   2. RWRachel Wolen13:50

      Yeah. Um, so the analytics agent is really, like, a way for me... You know, let's say I wanna understand what... how many, um, how many sites does Shirts have in its workspace? So I just ask it, like, a natural language question, and then it's basically going to go, and it's gonna write a query. So it's using tokens to write that query.

   3. AGAakash Gupta14:18

      Yeah.

   4. RWRachel Wolen14:18

      And then it... I've already authenticated, so it shouldn't go... It, it might... The way I've set it up is that my authentication... Yeah, so it's already authenticated, and it just basically spits back, "This is how many sites you have." So it's very much like it can act like an agent. If I asked it a more complex question, like helping understand the signup behavior over, you know, the last week and a half, um... Again, that's, like, proprietary data, so I, I probably won't ask that question on this podcast. But it will pass back to me and say, "Oh, well, this is how many visitors you have, and this is your week-over-week behavior." So it's almost like having a, a data scientist in my pocket. Um, and the way that I set this up is I actually have a analyst that I've set up that is, like, a, a Snowflake, uh... Sorry, I've, like, basically set up a agent that is a Snowflake agent that monitors, uh, different trends for me. And so then it, then it can basically go out, and this agent can go and, like, report meaningful changes and discrepancies.

   5. AGAakash Gupta15:24

      Mm.

   6. RWRachel Wolen15:24

      And so that is feeding into another agent that I've set up that allows me to go and analyze what is actually in my, uh, in my Snowpl- Snowflake repo.

   7. AGAakash Gupta15:35

      So how are you invoking these different agents, and what is the right way to organize these?

   8. RWRachel Wolen15:38

      Yeah. So one of the reasons why I like to keep these in separate windows is a lot of times, like for example, I'll, I'll go into the podcast prep 'cause this was kind of a fun one that I did for your, you know, for this. So what I did was you can basically pull your agent in, so that's, like, one way that you can invoke it. Usually, it picks it up. What I want it to do, right? So let's say I want you to prep me for a Aakash product growth podcast. Which agent are you using? And let's see if it, like, actually picks it up correctly. It should pick up this podcast prep researcher agent.

   9. AGAakash Gupta16:20

      Nice. So basically, just by having the context of the agent markdown file in your agents folder in the folder that you've opened up Cursor in-

   10. RWRachel Wolen16:28

       Exactly

   11. AGAakash Gupta16:28

       ... Claude can invoke it.

   12. RWRachel Wolen16:29

       See, it's using this. It's picking this up now, right?

   13. AGAakash Gupta16:31

       Love it. So it just invokes the agent just by having that markdown file there.

   14. RWRachel Wolen16:34

       Just by having it in-

   15. AGAakash Gupta16:35

       And you just keep them in an agents folder, it sounds like. There's not really much else to it.

   16. RWRachel Wolen16:39

       There's not much else. And that's where Claude generates it.

   17. AGAakash Gupta16:41

       Yep. Very cool.

   18. RWRachel Wolen16:42

       Um, so, like, for example, we'll maybe go and generate a LinkedIn post generator. But what I wanted to show you that I thought was pretty cool... So what I've also done beyond just... I like this trick-or-treating. That's fun. It's Halloween. Um-

   19. AGAakash Gupta16:55

       [laughs]

   20. RWRachel Wolen16:55

       What I wanna show you here is this is actually the output. So I run this as, like, a app. Um, so I know that I am in the middle of a podcast with you, and so I've kind of built this out as my calendar. Um, I also have, like, different agents that have outputs. These... This is the markdown file, so it's reading the markdown file, and then this is what it actually generated, and this was, like, the prep work that the agent did for me for this podcast, right?

   21. AGAakash Gupta17:22

       Pretty epic. I mean, it's pretty industrious. It's not just doing a little bit. It's going the next level.

8. 17:27 – 19:26

   Linear Ad Start

   1. AGAakash Gupta17:27

      This episode is brought to you by Linear. You know what's broken about product development right now? It's not the coding. We've got AI helping with that. It's everything else, the planning, the feedback synthesis, the endless context switching between tools. You're drowning in PM busy work while the actual product work suffers. Linear started as that issue tracker engineers actually wanted to use, which, let's be honest, is pretty rare, but it's evolved into something way more powerful, your entire product development hub. Their new AI features are a game changer for product managers. Product intelligence automatically surfaces insights from customer feedback and support tickets. No more manually combing through hundreds of reports. And their AI agents help draft PRDs, scope projects, and handle all those status updates nobody wants to write. Product teams from OpenAI to Vercel use Linear to build complex products at speed. There's a reason they call it magical. So visit linear.app/partners/aakash. That's linear.app/partners/A-A-K-A-S-H.

   2. RWRachel Wolen18:28

      So I have a, let's see, I have a video transcriber. So this is an agent where I will, like, pass several of your previous YouTube videos, and it will transcribe those videos, and it will pull out context for the, for our podcast session. So I f- look, I find it, like, not easy necessarily to read the markdown file, which is why I created the app for myself.

   3. AGAakash Gupta18:49

      Yeah.

   4. RWRachel Wolen18:49

      Um, and the app I'm, like, kind of constantly tweaking. Uh, I'll show you, like, one other version. I went to dinner last night, and I prepped myself for dinner, and literally all I did for that dinner was I gave it the... I, I gave it who... I, I gave it, like, my calendar invite of who was going. And then it, like, generated this amazing, like, this is who is gonna be there and, you know, what they c- what they have been talking about, and it went into their LinkedIn and, you know. So to me, this is, like, what an epic chief of staff would do.

   5. AGAakash Gupta19:19

      Yeah.

   6. RWRachel Wolen19:19

      Um, and I literally, like, all I did was I went into my dinner guest research, and I invoked the agent,

9. 19:26 – 29:16

   Building Agent from Scratch

   1. RWRachel Wolen19:26

      and it generated that.

   2. AGAakash Gupta19:26

      All right. So you just walked us through the analytics agent. The next thing is can you help show us from scratch how we would-

   3. RWRachel Wolen19:33

      Yep

   4. AGAakash Gupta19:33

      ... build an agent together?

   5. RWRachel Wolen19:34

      That sounds great. Now I'm gonna create a LinkedIn post generator, okay?

   6. AGAakash Gupta19:38

      All right.

   7. RWRachel Wolen19:39

      I have to create a lot of LinkedIn posts. I have a custom GPT, but I do think that this is, like, a new agent where it would be a lot easier if I just fed it a bunch of content. Um, and so I'll show you the way that I think about doing this. So all you do is go into agents. You manage your agent configurations. And then I'm gonna create a new agent. So this is literally just walking me through this, and then I'm gonna give it access to the whole project. You can give it less access, um, but I'm just gonna give it access to the whole project. And then I'm going to use Claude to generate that agent. And I want it to write a LinkedIn post and generate a meme image using OpenAI, uh, Image Gen model. Um-

   8. AGAakash Gupta20:26

      Maybe we should throw in there not to use an em dash so it doesn't give you away.

   9. RWRachel Wolen20:29

      There you go. Don't use an em da- Is that how you...? Yeah, probably. Close enough.

   10. AGAakash Gupta20:35

       [laughs]

   11. RWRachel Wolen20:36

       Is it em dash or is it em dash dash? I'm not sure.

   12. AGAakash Gupta20:39

       [laughs]

   13. RWRachel Wolen20:40

       Um, and then let's, let's also reference the materials I'm going to give you for what makes a great LinkedIn post. I'm also going to give you my best performing posts.

   14. AGAakash Gupta21:03

       Hmm. Nice.

   15. RWRachel Wolen21:04

       Okay. So that's what I give it, and then what it's gonna do is generate the agent. And then I'm going to point it to, uh, I, I kind of, like, grab some data to begin with. We'll get there in a second. It takes, it takes its time.

   16. AGAakash Gupta21:16

       [laughs] But at least it has fun verbs to let you watch along.

   17. RWRachel Wolen21:20

       I know. Doesn't it make you feel so good?

   18. AGAakash Gupta21:22

       [laughs]

   19. RWRachel Wolen21:22

       I, I, I like that it has personality. I, I feel more connected to it.

   20. AGAakash Gupta21:26

       And if you guys really hate it, just use Codex. [laughs]

   21. RWRachel Wolen21:29

       Well, I think Codex has, like, a, a time and a place. Um, okay. So I'm gonna create that and give it... I've been using Sonnet for these types of tasks. I think it does pretty well. And then we will make it... Let's make it blue 'cause LinkedIn's blue. And then so this is basically what it generated. You are an expert in LinkedIn content strategy, social media, blah, blah, blah. And then I'll show you, like, the full version of this. So it just created this file. So it created this file that is a markdown file. It's telling you what to do. You wanna analyze reference material. So you're gonna see I'm gonna point it at those reference materials, craft a compelling post, and then it's gonna generate a complementary meme image. Let's see if it works.

   22. AGAakash Gupta22:07

       Can I show everybody something cool?

   23. RWRachel Wolen22:09

       Yeah.

   24. AGAakash Gupta22:09

       Which I don't know if you know about yet. So if you two-finger click on the m- markdown file in the left-

   25. RWRachel Wolen22:14

       Mm-hmm

   26. AGAakash Gupta22:14

       ... you can actually open preview. So, like, kind of, like, right click-

   27. RWRachel Wolen22:18

       Okay

   28. AGAakash Gupta22:19

       ... on the markdown file name in the, in the left bar.

   29. RWRachel Wolen22:23

       This one?

   30. AGAakash Gupta22:23

       Uh, yeah. If you go all the way to the left-
10. 29:16 – 34:25

    Setting Up Your Org for AI

    1. RWRachel Wolen29:16

       well.

    2. AGAakash Gupta29:16

       And right before we get into that, the last thing I wanna ask you, we've just showed all these amazing workflows. As you talked about, as a leader, we have to motivate others and inspire others. How do you set up your organization to work this way? Obviously, you need to get everybody a Claude Code license. You need to allow people to access the MCPs for Snowflake and whatever else that might be needed. But beyond that, how do you really build a product organization that is at the bleeding edge of AI?

    3. RWRachel Wolen29:44

       This is a, a great question. So I think one of the things I would first assume is that your organization is going to be like every other adoption curve known to man. So you will have people in your organization that are the early adopters. You'll have the early majority, but you will also have the late adopters and the laggards, and then kind of everyone in the middle. And you wanna really figure out how to cater to all of those different people in your organization so that they can start to ascend the ladder themselves. So whereas I might be... I would probably put myself more in early majority, at least as, if it's hardware, but maybe with software I'm more in the, you know, the early adopter. I think that I have people in my team where I'm like, "Hey, I only wanna see prototypes," for example, "when you're gonna have meetings with me." And that's kind of created a dynamic where we spend a lot of time looking at the prototype, and they spend a lot of time investing in what that prototype, like, what that experience is like. Doesn't mean we don't have a PRD, but we've kind of, like, shifted away from a PRD in somewa- in, in a lot of cases, and maybe it's, like, a more evolving, you know, document, right? And so then, you know, when I kind of think about how do we train our team, we have, everybody has, like, access to Figma Make, for example.

    4. AGAakash Gupta30:59

       Hmm

    5. RWRachel Wolen31:00

       Um, is a very, like a much easier tool to sort of learn. And we've taken our design system and made that accessible in Figma Make. We also have a repo that is our design system that is accessible through Cursor. We just did a Cursor training, we did a Figma Make training, and we've done a couple of these builder days where we have people that are, like, the champions on our team who are maybe at the bleeding edge, but they are really there to, like, help walk people through getting over the technical hurdles. And so we're gonna do a builder day where, like, everybody has to demo something. Um, and it can be in any of those tools, but you do have to go a little bit outside of what you're comfortable with right now. And ideally, y- you know, what we saw when we, uh, went through the, that exercise was the first time we did it, we just did it in design, and we basically went from, like, nobody in design using Cursor to about 30% of the team using Cursor weekly. Um, now that's, like, kind of crept up a little bit because once you have, like, a base of people using it in an organization, you start to see more and more people leveling up. And then we're about to go do a second builder day, and it's going to be for product design and insights, which is, like, user research and data science. Data science is a different set of use cases for Cursor than product managers, than design. And so a lot of this is, like, trying to both have people who are champions that are, like, kind of bottoms up showing things off, and then also saying, "Here are some of the behavior changes that we expect." We are actually rewriting our, um, career ladder to incorporate this as, like, an expectation. We're thinking about, you know, so it's like you want people to be supported, but also you wanna create the right incentives inside of your team. Um, and then you also want to make sure that you're thinking through, like, w- well, are you just inserting AI for AI's sake? Are you going to get... At the end of the day, you wanna get to a better outcome. So, like, I'll give an example that happened to me yesterday. I was in a meeting, [chuckles] this is hilarious, uh, where this designer had put together an amazing prototype. It was awesome. It was, like, really, like, very future forward. It incorporated, like, a lot of new thing, n- new elements, um, around our answer engine optimization workflow. Uh, this is kind of the new AEO is the new SEO.

    6. AGAakash Gupta33:24

       Yeah.

    7. RWRachel Wolen33:24

       And what we -- [chuckles] but somebody, another, like, one of the directors of design on my team was like, "Hey, I was in this design review where somebody else had a prototype. It looked a lot l- you know, like some of the workflows you are building, and I want you two to go and, like, harmonize your two prototypes." This is a lot easier to do now than, like, being like you're, you know, so far down the, the product development life cycle and you're building something, and then you're like, "Oh, crap. These, you know, two workflows, like, don't work with each other." So I think it's, it's, it's really productive, but it, it's definitely, like, a different way of building.

    8. AGAakash Gupta33:56

       So many insights packed in there. If I were to synthesize for everybody, at the base of the pyramid, start with access. So we talked about Cursor access, Claude Code access, Figma Make access, then giving the MCP access for those tools. The second layer, supporting your team, whether that's training, builder days, bringing the people who are at the bleeding edge and helping others. And then we talked about getting the incentive structures right, so e- even changing your career ladder. That's how you actually create these AI-native product organizations.

11. 34:25 – 36:12

    Shipping AI-Native Features

    1. AGAakash Gupta34:25

       So once we conquer the productivity side, the other side is shipping AI-native features, and I think you have a really interesting story about eval. So can you tell us that?

    2. RWRachel Wolen34:35

       Yeah. So one of the things that I think is really fun about building AI-native products is so much is changing. So what you're seeing here is Webflow. Webflow is a, uh, website experience platform that is AI-native. And, you know, we -- this is a website that I built out for a website, um, a event planning company called Party Parrots, and what you'll see here is a set of components, um, and variables, uh, that are for this particular site. Now, we decided to build out an app gen product that uses those variables, uses those, the whole design system, as well as our CMS. That's, that's really, like, how we thought about differentiating. Now, what was funny is I was getting ready for this podcast. [chuckles] This product broke. [chuckles]

    3. AGAakash Gupta35:25

       Oh, no.

    4. RWRachel Wolen35:25

       And -- I know, exactly. And, uh, so obviously the product works, like, I've generated full apps. Um, but, you know, we're, we're getting -- we're about two weeks away from launching this product, uh, into the world, which is exciting. And also, we decided to go and change out the model. And so I was, like, patient zero going and, like, generating apps for this podcast, and what I realized, I, I kind of kept telling the team, like, "Hey, this is, this, the, the agent keeps dying. Why is it dying?" And we were trying to figure it out, and there were some other variables that we had, like, changed in the experience. And what we realized was we had changed the underlying model, and our evals didn't have enough coverage to fail when we changed the model.

    5. AGAakash Gupta36:11

       Mm-hmm.

12. 36:12 – 40:58

    Evals Story

    1. RWRachel Wolen36:12

       And I think that is one of the -- it, it, that is a new skill set for a lot of people is, is building evals, which are effectively test cases, um, for a model. And a lot of times you want a test case that is going to fail inherently, and that's really hard, but you also want test cases that you think will pass. And, you know, so each time you go and change out the model, you wanna see how the model do- the new model does against your, like, what I call, like, dream evals. And so in this case, like, we didn't -- we actually lacked the coverage. Um, and so we've been really trying to think through how do we instrument product leaders? How do we help product leaders? Again, this is one of those new tools that's part of the, the AI product manager, um, toolkit. So how do we teach PMs how to write evals? How do we teach them to have enough coverage?Um, so we've been working with our vendor, uh, we, we use BrainTrust-

    2. AGAakash Gupta37:06

       Yep

    3. RWRachel Wolen37:06

       ... and are trying to understand, well, what are the best practices across all the other teams that are out there? And, you know, where should we use, for example, synthetic evals, where we're, like, generating it using, uh, a- another model, right? And so I think that's been, like, a really interesting process for, especially for the... We have, we have a number of products that are AI-native products that we're building here. And but then it's also been very interesting to see, I mean, when you're building a code gen product, you know, like for us, we wanted it to be so simple that, like, literally all you did was give components and CMS collections and, like, a very, very simple prompt, and we wanted you to get something great.

    4. AGAakash Gupta37:43

       Sounds simple, but very powerful. [laughs]

    5. RWRachel Wolen37:45

       Very powerful. I mean, this is a chat kit app that actually works.

    6. AGAakash Gupta37:49

       Mm-hmm.

    7. RWRachel Wolen37:49

       So to me that, that was, like, the, you know, how do you build something that's differentiated in a AI-native space where it's maybe even noisy? We still think that's very possible, and I think that's, like, been a very interesting process for us to, to go through.

    8. AGAakash Gupta38:04

       So the next lesson is you need to choose features that are actually related to your strengths. How did you guys decide, and how do teams decide what AI features to be putting on their roadmaps?

    9. RWRachel Wolen38:15

       I think that's a... You know, part of it is, like, number one, if you see a trend in market, do you think that trend is applicable to your customers? So I'll-

    10. AGAakash Gupta38:23

        Mm

    11. RWRachel Wolen38:23

        ... I'll share, like, another app that, and just kind of like walk through that. So in our case, our strengths are really helping our customers bring visitors to their front door, right? And so we think of our CMS, our, uh, which is a, um, which is basically like a collection, uh, of database items that then get rendered for search engines and answer engines. An answer engine is like ChatGPT to discover. So this is our core competency. What we saw was like, hey, we think there's a lot of people who, for example, don't necessarily want to have an app off to the side generated by, you know, Lovable or v0. That's more like a prototype. We want, we, people wanna generate, like, production apps.

    12. AGAakash Gupta39:01

        Yeah.

    13. RWRachel Wolen39:01

        And so what does it mean to have a production-grade app? And so I'm like, well, it looks like your brand. That, I think that was, like, one of the, the first principles. Um, it uses your, you know, it, it uses, like, your, your CMS. So this actually uses your CMS to generate these. And then it can integrate into your workflows, and so we've really focused on how is this natively integrated with everything else and still simple enough for our customers. We, we have kind of a wide variety of different types of users, everything from a designer who maybe might be technical but not a developer. We have developers on our platform, and we have marketers that are, you know, like content marketers or even, um, performance marketers that are not very technical. And so we wanted a product that could actually cover that gamut and f- like, use our production-grade hosting, benefit from all the security capabilities we've built into Webflow Cloud, et cetera. And so a lot of that is, like, we've built a lot of the scaffolding around this, and we're like, "Okay, we're not just bringing a, uh, you know, a coding agent to market. We're bringing a way for you to prompt an app to production." And that, we believe, is, like, quite a differentiated experience than what you would get out of the box from another coding agent.

    14. AGAakash Gupta40:20

        Yeah. I think this is extremely powerful. As a product leader, you need to be leveraging the latest technology. You can't see a trend, [chuckles] like all this stuff we just demoed with Claude Code, and not think about, "How could this be brought to my product?" And I love the thinking framework you've given us here of, well, what are my strengths? In Webflow's case, you know, building production apps, having an incredibly large enterprise backing of users, being able to do it on your hosting, being integrated with the CMS. For people who don't understand CMS, content management system, this is, like, where you're actually publishing all your blogs and your pages and the content on them, and then leveraging that to build a product. I think this is the art of building amazing AI-native

13. 40:58 – 44:00

    Distribution First Mindset

    1. AGAakash Gupta40:58

       products. The final lesson is you need to think about distribution. How do you build products with a distribution-first mindset?

    2. RWRachel Wolen41:04

       Yeah. It's a great question. So I think that if you think about the different waves of innovation that have happened, um, you know, each of those waves has had a different distribution mechanism that product managers, um, have had to learn. And so if you think about the first wave, that was really the internet, and then you needed a way to find websites. So people learned how to optimize search engines. Search engine optimization, there's a lot of dark art to optimizing your website for keywords and keyword stuffing back in the day. Um, but-

    3. AGAakash Gupta41:37

       It still works. [laughs]

    4. RWRachel Wolen41:38

       It still, it still works sometimes. Uh, and so that was kind of like phase one. Phase two was you're like, "Okay, I also want to launch a mobile app," and mobile apps also have app engine optimization, right? So people did, had very similar tactics to get to the top of the iOS store, right? And so that, again, was a very specific set of tactics-

    5. AGAakash Gupta42:03

       Mm

    6. RWRachel Wolen42:03

       ... to get distribution for your product. Wave three was social, so you were building for virality. You're trying to get people to discover your product, to come back to your product, to share your product, and on and on and on. That, you know, worked on... It still works on many different platforms. And so these are all still viable distribution paths for some segment, but the next wave is really going to be around how do you get listed in answer engines? How do you get your brand recognized in answer engines? How do you get your, you know, if you make a product change, it's like a major product moment, how do you get that answer engine knowledge to swap out? And so a big part of that is going to come through your website, understanding that you need to feed, you know, a FAQ to your website and get that updated, and that is something that needs to be part of your product process in order to get that information updated. Um, or you can use a platform like Webflow that does it automatically. Uh, you need to really think about-How are you going to create apps that showcase or maybe highlight part of your experience as, you know, ChatGPT apps start getting adopted? And inevitably, there's going to be all kinds of agentic experiences around these agentic browsers, uh, that will also want to interact with your app or with your website or with your e-commerce store, et cetera. So really thinking about... I mean, and that's gonna be a huge opportunity for growth, right? That's gonna be an unseating of how people discover things. And so the more you can understand w- how people's discovery is shifting, the more you can actually drive growth to your product.

    7. AGAakash Gupta43:44

       Yes. I think that this is the final thing that so many PMs, they maybe outsource to product marketing or outsource to their execs, but it's actually worth thinking about from the very beginning so that you make it a part of all your plans. So that concludes our masterclass in AI product leadership.

14. 44:00 – 45:32

    Outro

    1. AGAakash Gupta44:00

       We started with AI productivity. We walked you through how to be an IC CPO, how to use Claude Code and Cursor, how to build agents, what agents Rachel is building, including her amazing chief of staff team of agents, and how to set up your organization. Then we talked you through the three most important lessons on shipping AI-native features, having great evals in place, continually iterating on them, playing off your strength, and thinking about distribution first. This is the roadmap to becoming a great product leader, a top 1% product manager. You have to embrace these tools. Rachel, thank you so much for dropping all the sauce.

    2. RWRachel Wolen44:32

       Thank you so much, Aakash. It was great to be... to come on, and just really appreciate it. Thanks for letting me share.

    3. AGAakash Gupta44:38

       And if you guys didn't know, I write a paid newsletter of which Rachel has been a subscriber for a really long time, which is really, really special. I only found that out in our pre-call recording. Check that out as well if you want. Check out Webflow for their amazing new app builder, and we'll see you in the next one. Bye. So if you wanna learn more about how to shift to this way of working, check out our full conversation on Apple or Spotify Podcasts. And if you want the actual documents that we showed, the tools and frameworks and public links, be sure to check out my newsletter post with all of the details. Finally, thank you so much for watching. It would really mean a lot if you could make sure you are subscribed on YouTube, following on Apple or Spotify Podcasts, and leave us a review on those platforms. That really helps grow the podcast and support our work so that we can do bigger and better productions. I'll see you in the next one.

Episode duration: 45:42