Transcript: She Built an AI Product Manager Bringing in Six Figures

Timestamps

Introduction: 00:01:02
How the groundwork for ChatPRD was laid: 00:02:15
Why building solo—with AI—is faster and cheaper: 00:12:38
Claire demos ChatPRD live: 00:14:48
Testing the document editor feature in ChatPRD: 00:22:44
How ChatPRD is baked into Claire’s workflow: 00:26:13
Claire’s ability to build a side project—pre-AI v. post-AI: 00:33:13
The future of product management: 00:36:22
How Claire drafted a product strategy during her 22-minute commute: 00:43:50
Using AI as a tech-forward parent: 00:45:55

Transcript

Dan Shipper (00:01:02)

Claire, welcome to the show.

Claire Vo (00:01:05)

Hi. I'm so excited to be here.

Dan Shipper (00:01:06)

I'm really excited to have you. So, for people who don't know, you are the chief product officer of LaunchDarkly, a feature management and experimentation platform. And you are also the founder of ChatPRD, which is an on-demand chief product officer that writes and improves your PRDs. And we met at a sort of angel investing retreat and I was doing bits about AI, and we just got into this really great conversation. I feel like you're in this really interesting intersection of leading a business as a chief product officer and then you're also building this ChatPRD side hustle that is built with AI and uses AI. And so, I think you just have so many ideas going on in your brain about what the future of this looks like and so I'm just super excited to have you on.

Claire Vo (00:01:56)

Well, I'm excited to be here. It's something that I think about a lot. And as I was preparing for this conversation, I took a moment and reflected that I use a lot. So, I'm excited to just share a little bit about what I think and what I'm building.

Dan Shipper (00:02:05)

That's awesome. So, why don't you give us a little bit of a background on ChatPRD because that's the thing I'm most interested in? It seems so cool.

Claire Vo (00:02:15)

Yeah. So, like all good workers—well, maybe not like all good workers—I'm trying to work myself out of work. That's the ideal way to do things. And so when ChatGPT and some of these tools came out, I was all over it. It did not scare me. It felt like magic.

And I spent a lot of time as an executive leader. I lead product and engineering organizations—many hundreds of people, pretty large responsibility. But I still have, yeah— Believe it or not, more than meetings and hiring great people to do, I actually have to output work. And so, with a calendar like mine and the demands of a job like mine, anything that can be a help is very welcome.

So I started using ChatGPT to help me basically write product strategies and product specs. We were a fairly, I’d call it scrappy team at my previous company and occasionally I would PM our more technical products. And so something would come up, it would be pretty complex, it would span product and engineering, and I would raise my hand and say, hey, I'll write the spec for that. I think I have a sense of what we need to do.

And there's this very specific example of us building a pretty complex and custom data audit tool. And I raised my hand in the meeting, I remember, at 10 a.m. And then by 2:00, I had this full five-page spec and my team was like, what just happened? How did you do that? Because you've been in meetings all day. What was that? And it was because I had, over the course of months, sort of prompted ChatGPT into a place where I could really work with it in a pretty rapid fashion to get high-quality outputs. That wasn't just going to happen with just using kind of plain ChatGPT, like GPT-4 at the time, I think.

So when the GPT Store came out, I thought, okay, I'm just gonna drop my— My joke is it is just a prompt, but she's my prompt. And so I dropped my prompt into the GPT Store, got the great name, ChatPRD, which I think is just— It’s good stuff. People love it and shared it with my team and I was like, you all can use this if you wanna know how I do it. And they all loved it. And so I was just kind of joking around and to my husband, I said, we should just buy the domain on it. Of course it's gotta be dot AI, so I spent my $60 or whatever, there's a premium on these dot AI domains and bought ChatPRD.ai, and just put up a newsletter signup form and a link to the GPT. And so many people started using it. And still, with the GPT Store, I think it's early, early days. So even though it's getting a lot of use, there was no monetizing things. And I'm sorry, I'm the kind of product person that needs to make money off the things that they build.

So, over Thanksgiving last year, between hosting the kids and doing stuff when they were napping, I dusted off VS Code and I was like, yeah, I think we can build this. And so I built it over the course of Thanksgiving week with my kids home while they napped. We can talk about all the ways I used AI to make that really, really easy and launched it, I think, on November 28th. So she's six months old. And just put, I don't know, $2 a month. I was like, I don't know. Who knows what somebody is going to pay here. And people started buying. And then I started to add more features and raising the price and people started buying. And now I have, I think, 10,000 users on the app that I host, over 50,000 chats have been done on the GPT—somewhere between 50,000-100,000 because I think that's where they bump you to the next tier. And I started to build out a lot more than just chat features on the product. So it's been really fun to build, and I think I have 285 ChatPRD chats in the last six months. So I use it a lot.

Dan Shipper (00:06:16)

That's wild. There's so many things I want to dig into there, but I want to just summarize the process that led you here because I think it's a sort of generalizable thing. I actually have something similar in my own life that I've done with this where it seems like you used it yourself and you built up a very complicated prompt that reflected your worldview and skills and your process for making PRDs. Actually, before I keep going, can you define a PRD for people who don't know?

Claire Vo (00:06:48)

Yeah, I now say ChatPRD is an on-demand chief product officer that helps you with your product work. Because we've really expanded past PRDs, but a PRD is a product requirements document. It's generally a doc or a spec that defines the problem space that you're going after when you're building a new product, kind of what users want—if you have one user or multiple users of the product—what are their incentives, what do they want, what are their goals, what are their non-goals, and then kind of details out some of the features and capabilities you would need to make this meet the needs of their goals. And then a lot of teams and companies use it also to outline things like tracking plans, security risks, technical considerations, marketing campaigns. So it's sort of the source of truth for when you're building something new as a product, whether it's a very small feature or a brand new company, sort of the written record of what that's going to be.

Dan Shipper (00:07:42)

Okay. That's great. So basically in the process of doing these yourself, you realize ChatGPT is really useful. You create these prompts that help you build good PRDs. And then you realize, oh, this is sort of complicated for me to prompt manually all the time. And then you start just making a GPT and then you share it with your team. Your team loves it. And then you're sort of off to the races like, okay, I'll have it on the store. Maybe I'll build my own version of it, but I think that's the sort of repeatable process that I think what a lot of people are finding is, off on your own explorations, make something awesome, make some awesome prompts, and then you can sort of productize that.

Claire Vo (00:08:20)

Yeah. And I think the key that we're not talking about is how do you distribute it? Because unless you are maybe featured in the GPT Store, which can get you a distribution, you really own your own distribution for even the GPT Store or an app. And so I was lucky enough to have a good number of people who follow me for product content. So it was right in the wheelhouse of who might find it useful to use the product. And then it creates that virtuous cycle of word-of-mouth distribution, but I do think distribution still is a challenge for anybody kind of going down this path.

Dan Shipper (00:08:54)

Yeah, that makes total sense. And so just to back up to sort of the beginning of the story, tell us about what you were using it for initially. What were the sort of initial prompts that you were working on? What were the tasks or the mini-tasks within making these documents that were actually helpful for you?

Claire Vo (00:09:14)

Yeah, so basically we have a PRD template. So we have, every time you're going to build something, here's what we just generally expect you to outline and write. And then as people write those, then we have, at my old company, we call it PRDR, product requirements document review meeting, where we actually debate both asynchronously in comments and then live, sort of the thinking in this document, and then it becomes a source of truth for what to build. So we had this template and PMs, including myself, would once a week sit down and be like, I got to pick up that next project and sit down and start with an almost blank slate template, and go, okay, I got to write this whole thing out. And they end up being somewhere between a mini-PRD would be, just like a one-page up to seven or 10 pages, just depending on the complexity of the product. And a lot of times you're doing things, like it needs to have rules-based access and these very nonspecific but functional requirements need to be outlined and, if you forget them, then engineering goes off the rails and you get a product where you forget to build a whole bunch of table-stakes features. And so what I would do is I would paste it and I was like, this is our template. This is the type of company I'm working at. And then just word vomit, here's the product that I think we need to build and why. And it would scaffold out, then, all those sections combining the context of the template, what I said about the business, kind of high level what I've said about the product I want to build. And the thing that really was the aha moment for me was not that it could do that, obviously could do that, it was that it thought of functional requirements I had, I would have truly forgotten it and I think I'm pretty good. I've been doing product for 20 years. I am pretty good.

And when I was building this audit feature, I was focused so much on the data we needed to ingest and how people needed to audit. And then when I was generating this PRD via AI, it said you need filters. And I was like, of course I need filters. Obviously I need filters. And somebody probably in a meeting would have said, Claire, you need filters. But it just put it in the doc. And it was just one of these moments where I was like, oh, it's a little buddy that I can have that can just round out the edges of my thinking, make me a lot more efficient and remind me of things even that I need to include where my fallible human brain fails me.

Dan Shipper (00:11:36)

I love that. I think that's really common too. Some of the stuff that you learn from, it's not stuff that you wouldn't already know. It's just you might not have thought of it. It finds the simple things that you're missing and that is sometimes so valuable, especially if you're working alone. You have a team and stuff like that, but building ChatPRD where I think it's you and maybe one other person part-time. It's sometimes hard without a buddy to remember that stuff and ChatGPT is that. I'm really curious, to get into how you're using it, do you want to show us some of your historical ChatPRD chats?

Claire Vo (00:12:17)

Yeah! I can share my screen. Okay. So, this is ChatPRD. And some of the things that I think are really unique compared to the GPT Store is how, one, not only I've customized it, but you can actually customize it to yourself. So you can tell—

Dan Shipper (00:12:37)

Wait. Can I pause you right here for one second? This is really beautiful and well-built. How is this built?

Claire Vo (00:12:43)

Oh! I did it. I mean, this isn't magic. If anybody's building stuff right now, I'm going to be like, it's an XJS app, mostly Tailwind. I have good taste. I don't know. I think one of my other learnings here is that it is so cheap and easy to build high quality web apps right now. And as he's built web apps many times in their lives and run very large engineering teams that build web apps it's actually not that hard to build something pretty phenomenal. It's a database on a website. So thank you. I appreciate that. I try to put love in it. And we can talk about how actually I think building solo does increase quality and velocity in many ways because you're short-circuiting the loop between product engineering and design in a way where, I think in a traditional team where you have multiple people playing that role, you lose fidelity in that vision. And so one of my interesting ideas is with something like ChatPRD, like plus Copilot or like a Devin, do you actually collapse and focus on the ability of one person to build something? And does the quality actually go up because you're not deciding by committee, you're deciding kind of one person.

Dan Shipper (00:14:00)

Having everything integrated— It's the same thing with the multimodal models. It's like a lot of the stuff gets better when you integrate the image recognition and the video into the intelligence portion. And it's like, yeah, integrating development, design, and product thinking into one person is going to— you’ll have a faster cycle. So you're going to be higher quality. I think that's really cool. And did you use ChatGPT to code this or did you code it by hand?

Claire Vo (00:14:25)

Of course I used ChatGPT! I mean, look, I can write some code, but it's much easier to say— I mean, I take a PRD from ChatPRD and like, hey, this is the component I need to build. Can you just scaffold it out for me? And then I can play with the edges on design or functionality and that kind of stuff.

Dan Shipper (00:14:45)

That's great. Okay, so, let's go back to before I interrupted you. So, you're in settings.

Claire Vo (00:14:48)

Yeah. So you can customize your profile. So one, you can tell it a lot about yourself—who you are and what you do. It remembers that. So it has memory. So unlike— Well, I know ChatGPT has memory now, but a lot of times previously when I was building this, I'd be like, tell me about your role in product, tell me a little bit more about your company. And so we're just stabilizing that information. And then, as I said, most companies have a template. I have a pretty good one that, I think, I use as a default, obviously, because it's mine. But you can also add your own template of PRDs that say you have a TLDR, which I always do, a short summary of the product doc. And then you have functional requirements that are super important and “put functional requirements in X, Y, Z format,” whatever it is you can actually use a custom template here. This is something that's just really useful for. People as they're using this in the business context to get consistency and, again, not have to manage that prompting. But what I use it a lot for are two things. I use it to— Well, maybe I use it for three things. I personally use it for writing PRDs for ideas I already have or brainstorming features on the roadmap. And then as a product leader, I use this, help me improve an existing PRD to coach my team on how they can improve their PRDs. So those are sort of three things, but let's just talk about, help me write a PRD. So one of the things that I've been working on is a teams feature for ChatPRD. So right now, ChatPRD is sort of like in single-player mode where individual PMs can use it, but we've gotten a lot of feedback that people want a teams version of ChatPRD. So I'm going to say, help me write a PRD. I'm going to be really lazy and say, “teams plan features for ChatPRD.” Oh, okay. So then I just click go and—

Dan Shipper (00:16:47)

Do you normally do that level of detail? Or are you normally writing more stuff when you're actually doing it?

Claire Vo (00:16:52)

No, I'm pretty lazy. Because again, I’ve prompted ChatPRD to ask for stuff if it doesn't have enough information to do a good job. So you can see, okay, well, that's great. But what do you want? And so I say, “I've gotten a lot of feedback that teams want to use ChatPRD together. They have a few priorities: shared billing, shared templates and stored company context knowledge, sharing chats and documents with each other. If you have other ideas, please share them. I think we can grow to 100 teams accounts by the end of the quarter.” It's very, very high level, so it's going to be great. Here it is. And so gives me a problem statement, business goals, user personas, team lead, user experience, like how you might onboard team members, how shared billing and all this stuff works. A narrative, which is a little thing that I've injected here, which I feel like PMs are very bad at pitching their story. So, I've added that in success metrics, technical considerations, milestones, and sequencing, etc. Okay. This is an okay start, but I actually feel like the user stories are really lame. So I'm going to say, expand out the user stories, focus on a few personas: billing lead, team lead, admin, individual, team, users.

Dan Shipper (00:18:45)

Before you hit enter, I'm just sort of curious, what do you do as a product leader— When you're reading these, what stands out to you as the user stories are sort of lame?

Claire Vo (00:19:00)

It's just very, very high level. So they're just, “yeah, obviously, but then comma how?” And so I just think, yes, I want to share chats and documents with my team so we can collaborate efficiently, but let's get much more specific. Actually, let's get a little more detailed. Let me see. “... a lot more detailed outline features and sub features that might be useful to all these personas.” That's great—

Dan Shipper (00:19:30)

I think I'm asking, because for me, sometimes it's hard, seeing something from ChatGPT or honestly from people that work for me. I can kind of tell that it's kind of high level, but it's hard for me to— My eyes can just sort of glaze over it. I can skip it and be like, yeah, that kind of makes sense. I want to know what's in your brain to be like, okay, no, no, we gotta push in there.

Claire Vo (00:19:55)

Yeah. Yeah. I think it's specificity and I think this really shows itself up in the user experience. Now we've gotten something a lot more detailed onboarding team members where, okay, we're going to invite them via email. There’s a welcome guide. There's a secure link. There are roles and permissions here. Actually the roles and permissions that you need to build. So this, to me, feels much more specific. I'm like, oh, okay, I can hand this to an engineer now. And instead of saying we need roles and permissions, it says we need these three roles and permissions.

Dan Shipper (00:20:26)

Yeah. You can see it.

Claire Vo (00:20:30)

Yeah. You can. And so here's where I think ChatPRD, the app. gets really interesting. I think, “This looks great. I want to build this all in three weeks. Today is June 5th. Update the timeline.” So then the only other thing is the milestones. And so then it's going to regenerate this document, including all the updated user stories, which we have. So now we have these more detailed user stories, which I'm very happy about. And then it should generate more milestones in a more detailed way, and I tend to be very aggressive, so I give it like, I want to do this in two weeks or three weeks or whatever. Most teams tend to know how quickly they want to do things, but have a hard time breaking down those milestones. And so I find I use milestoning and sequencing pretty frequently here as a structure or scaffold for how we might do a milestone. So here, look, it's like initial development, billing features, 10-14th, 13-17th beta testing and launch. So it's got this nice, okay, can I hold myself accountable to that? Can I build that into Jira? And then this is where we go beyond sort of what you could do in a GPT Store, which is we have this idea of “save as a document.” Look at this nice little solo where— It's slow. Sometimes docs are incomplete. I've apparently pushed, I've talked to OpenAI about this, like their actual team, and I've pushed the edges of some of their function calling. So we're just going to see. We're going to wait for this to load. And while it loads, I'm wondering if you have any questions?

Dan Shipper (00:22:11)

I always have questions.

Claire Vo (00:22:12)

This guy takes forever.

Dan Shipper (00:22:14)

That's very cool. One of the things I was going to ask you is the decision to have this be a chat feature instead of a document that you are progressively filling out, a document editor type interface. What did you think about that choice?

Claire Vo (00:22:31)

Well, it's actually both. So if we can get this loader to— If we can get the robots to spit back out— I actually think multimodal is the way. Okay! That wasn't too bad. Look at that. So documents have been saved. So if I click Document Save, now I have it in a document format, I can actually toggle back and forth between Chat Mode and Doc Mode. This is the magic. And then—

Dan Shipper (00:22:59)