In this episode of 'The Next Next,' host Jason Jacobs interviews Paul Heayn, a product manager at Peapod Digital Labs and former colleague at Runkeeper. They explore Paul's extensive use of AI in his work and personal projects. Paul shares insights on how AI has transformed his workflow, the tools he uses, and the challenges he's faced. The discussion covers the potential of AI to automate tasks, the importance of learning to code, and how AI might reshape traditional roles in startups. Jason and Paul also touch on the future of AI in personal software development and offer practical advice for integrating AI into daily workflows. The episode concludes with a reflection on the challenges and potential of AI, along with an emphasis on continued learning and experimentation.
Exploring AI Integration in Product Management with Paul Heayn In this episode of The Next Next, host Jason Jacobs welcomes Paul Heayn, a product manager at Peapod Digital Labs, who has been integrating AI into his work. The conversation covers Paul's professional background, his journey with AI, and his hands-on experience with tools like ChatGPT and N8N. They discuss the challenges and potential of automating workflows, the future impact of AI on roles within startups, and how to effectively integrate AI into existing processes. The episode serves as a practical guide for anyone looking to incorporate AI tools into their professional workflows.
00:00 Introduction to Today's Episode
02:22 Paul Heayn's Background and AI Journey
03:32 Exploring AI Tools and Applications
07:48 Challenges and Learnings in AI Implementation
08:38 Future Prospects and Personal Reflections
17:18 Navigating AI Tools and Resources
21:02 Debating the Best Approach to Learning AI
27:16 The Programming Genie Concept
27:44 Challenges in Character Development
28:22 Automation in Various Professions
28:52 Building Personal Software Solutions
30:19 Future of Vendor-Sold Software
31:24 AI's Impact on Job Roles
34:56 Choosing the Right AI Tools
42:16 Practical Coding Tools and Automation
46:11 Final Thoughts and Future Plans
Jason Jacobs: On today's episode of The Next Next, our guest is Paul Heayn. Paul's a product manager at Peapod Digital Labs, and I've known Paul a long time. I actually worked with him at Runkeeper back in the day, from 2015 through 2018. At any rate, Paul's been using AI a bunch in his day job, and he's also been messing around with it quite a bit, nights and weekends.
Has been following the newsletter. He resonated with some of the things that I was talking about and thinking about, he reached out to me and it turns out that a lot of the things I'm trying to understand better are things that he's been trying to understand better in his hands. Have been dirtier than mine.
So this episode is essentially talking through what he's been doing, how he's been going about it, what he's been learning so far, and also his advice for someone like me, who's trying to figure out how to incorporate these tools into my workflows as well. Definitely an AI one on one type of deal, but I found it useful and hopefully you do as well, but before we get started.
I'm Jason Jacobs, and this is the next, next, it's not really a show. It's more of a learning journey to explore how founders can build ambitious companies while being present for family and not compromising flexibility and control, and also how emerging AI tools can assist with that. Each week, we bring on guests who are at the tip of the spear on redefining how ambitious companies get built.
And selfishly, the goal is for this to help me better understand how to do that myself. While bringing all of you along for the ride. Not sure where this is going to go, but it's going to be fun.
Okay, Paul Heayn, welcome to the show.
Paul Heayn: Thanks for having me.
Jason Jacobs: Thanks for coming. Yeah it's interesting, we we worked together at Runkeeper a long time ago, and I've been to my wife's chagrin living out loud here as I'm figuring out what's next, and I guess some of the things I've been talking about are some of the things that you've been thinking about and tricking with I was excited to I'm excited to hear from you and catch up and excited to hear that you're interested in some of the same things as me.
Paul Heayn: Yeah. And I think that's definitely true just on both the where you are in life and then also like, how do you use and learn about AI in general? So it's definitely definitely a collision of kind of paths.
Jason Jacobs: Paul, for context for listeners, maybe just set the stage, talk a bit about your professional background in product management and also what led you to start tinkering with AI? How long ago and where are you today?
Paul Heayn: Yeah so I got my start as a PM at Runkeeper and there I learned a bunch of building products that people love and working with UX. Going through the process and learning about the different metrics and how to just build a holistic product with an entire team.
Moved on to another
Jason Jacobs: Paul, remind me, what were you doing before you were doing product management?
Paul Heayn: So I was I was a web developer using cold fusion, which I don't know if I don't think anyone knows that. But it's now Adobe. And it's used only in like DC and San Francisco. But it's like a script language. And then I went to business school and wanted to be closer to the tech, but not Okay.
Be doing the actual coding. I wanted to be a little bit more strategic. And so I went to business school, found out about product management and loved it. And then I went to Runkeeper and learned there and ran from there.
Jason Jacobs: Got it. And so talk a little bit about what led you to AI when you started tinkering with it and how far along you are in your expedition.
Paul Heayn: Yeah, so I Started tinkering when ChatGPT, actually a little bit before ChatGPT kind of blew up. I heard that there were these different uses to use LLMs to try to basically create or automate some of your emails. I was like, oh, that sounds interesting. And it could write stories and stuff like that or change voices.
Like the tone of voice in how you write. Oh, that's interesting. That's pretty cool. But then like when I, you tried chat, QBT, there was this kind of like moment of let down where it's just Oh, this is all it does. And I didn't really get it put it on the back burner and then try it again a couple months later use it to code a little bit.
And I was like, Oh, this is super interesting. But I don't know necessarily what to code. And then. Earlier, like a couple months ago, one of my product analysts who's amazing, he left and I was like, how am I going to continue to do all the work at my current job? So I tried to build a bunch of different, what I thought were agents to replicate a lot of what he did.
And that's where I went down the path. Also like managing how do you deal with kind of corporate confidentiality and dealing with a lot of these things that you have to when you're not like in a startup you have to follow certain procedures. And then talking with you, I realized that Oh, there's a whole nother world.
And I think there's like an onion of AI where like you start unpeeling it and you're just like, Whoa, you can do a lot with this. But then there's like the step back of Oh. Should you do all of this with LLMs or should you program some of this or, um, have a mix of both.
And so like now I'm of the opinion that you want both like in there, like you want a reasoning agent to, to figure out what tools to pull off the top and really dive into accomplish your goals. And you want to automate it as much as possible, but then you also want to make sure that you're checking its work and it's not hallucinating on you.
Jason Jacobs: So how did you make out in terms of trying to, so did you, were you trying to use AI to replicate your tasks or were you trying to use AI to replicate the tasks of the business analyst who left?
Paul Heayn: So I was trying to replicate both, it was both, so both what he was doing so any kind of like user story writing or like PRDs or sorry product relate requirement documents or really like sprint review, like presentations, a lot of decks and stuff like that. Trying to automate a lot of that from connecting with JIRA and really trying to figure out like, how can you just remove a lot of that I don't want to say pain, but the stuff that you have to do as a product manager, like the communication parts that you have to do.
And that's just, that are just like time sucks that the stakeholders need.
Jason Jacobs: Huh. And so how did you figure out What to build using AI and how effective were you at building it?
Paul Heayn: I just started small and how I went about it, because again, this was. Using it in a corporate structure was, I would basically think about was a small piece of a tool that I could like or an action that I could automate and use AI to transform the data. And so it would basically I have a very bare bones, notetaking application, which will like just summarize it and summarize the meeting and then I can round up all of the other meetings and get context and search for, different contexts across different meetings and have PRDs spit out from that or have stories spit out from that.
And like, How I did this was basically like, it's nothing like super complicated, but because I had this corporate kind of structure, I had a Python based. Like application, which connected to our internal LLM. And then I would use perplexity to ask it questions and copy the code directly in, but all the API keys were in a separate file.
So I was able to like just basically copy everything in. It's very slow, but pretty effective at learning how to actually build this and understand how the code was working. So it was like learning at the same, like relearning how to code at the same time.
Jason Jacobs: Huh. And what were the biggest learnings or surprises through that process?
Paul Heayn: I think it was the The biggest one was when you realize that you could ask it to debug itself And like it seems common knowledge now, but like you just spit it back and be like, hey, I got this error Can you help me through this? And it's oh, yeah, like I made this mistake and because it has the context window It has the memory and there it knows the code and we can go back and give you the entire code back Then you just copy and paste that and bring it over so I think that was one of the aha moments that like I changed from like when I was learning that Oh, it could code to like, Oh, it can like actually be like an a whole development team that's helping you out.
And I don't mean that as in this will replace developers. I mean that this will help you and augment what you do in development and how do you test.
Jason Jacobs: Huh. And so going forward, do you feel like you've maxed out what I can do for you in your day job? Or are do you see lanes where you can continue to do more with these tools over time?
Paul Heayn: I, I think that I haven't maxed out at all. I have built a number of. Semi autonomous agents that, or semi autonomous agentic code. Which basically means that I am still the person who is running the scripts. And it's going off and doing some agent like activities. But it's not completely automated, and it's not like something that I have triggered yet.
So I think that's like, From talking with you in the past couple of days, it made me like dive into the. Agent space more so than I have looked in the last couple months and just figured out like what are the different One frameworks. What are the different? tools that are available and also like how do these things all function how they all the work how they all work together and What are the pros and cons between a UI wrapper of a framework versus like just building the actual agent in code So I think that there was a number of different learnings that I was just like trying to figure out and trying to understand and even like you and I we were approaching agents differently like I imagine like From what I've read, you don't necessarily want to dive into the code and be code heavy.
I imagine you want to be like integrating with like tools that have that are like no code or low code. But you don't necessarily want to get in and build your own scripts or maybe I'm putting words into your mouth.
Jason Jacobs: I want to be in my zone of excellence and delight, right? And I want to build big shit. And and so what I'm working through is how to take the steps that will help me to achieve both of those things. Historically, I've been a mouth. I haven't. Use my hands much, right? And and and I've typically tried to understand where the North Star is of what we're trying to achieve and then what skill sets are required to go and achieve it and then go get the best people for each skill set to go bring it to life.
These tools are supposedly Changing a lot in terms of what each person can do. But is it, does that change the calculus of how I should think about team building? I don't know yet, and I'm really wrestling because, for example, it's helpful to get in and use the tools, and I've been doing it and each time I'm getting a little further, but still, just to get something basic done, it's alright, I'll go copy the API key from this place, and then the URL from that place, and then and then get the the text editor from this place, and terminal, and I'm still learning about what all these things are, and then it's alright, and then You know, publish, and it's like, error, right?
And it's I can't publish anything without an error no matter how small it is, and no matter which service I use, and then I just end up chasing my own tail around for an hour or two, and then I get frustrated and bag it, and then I try again, a couple weeks later, right? And that's been my journey with AI so far, which is, which I wouldn't call productive other than to, to experience first hand maybe some of where it's at for people like me.
Paul Heayn: Yeah, and I think you bring up a good point, which is like, They're you're building AI products either for yourself, which like don't have to be productized They don't have to be like, you know hardened or anything like that They can just live in your folders and you know you're running them in your own local environments versus having a local server, which is a little more complicated which is like Then the next step is like publishing it online and you have to deal with different dependencies and now you have the local server and you're like hosted server and it gets a lot more complicated.
So that's why I've focused on like just building the things that I'm building, like just locally just, I would say like the one suggestion I would say for you is don't worry about publishing it, get it to work, get it to like function how you want and like figure out like What is the process that you like?
What is your lane of excellence? Is it that like you want to be in the code and you want to learn that stuff? Or is it that like you want to get a product to out to people? Cause I think that's, those are two different things like that you go about and figure out like which path
Jason Jacobs: I think the tricky thing for me is that where I'm starting is around this kind of learning journey, right? And and in its earliest phases for as far as the eye can see, it's not a business, right? And so there's something empowering about, for example with MCJ in the early days. Every podcast episode, I would send to an editor, and I would pay out of pocket, and And the editor would turn around, and it would sound nice, but it would cost money.
Every time, right? And actually, each episode cost Orders of magnitude more than a month of the script for editing All You Can Eat, or maybe it's not All You Can Eat, but I haven't hit the cap yet, and I'm Editing a ton. And I'm doing it all myself. So it's cheaper, it's faster, I have more control.
Like, why wouldn't I use Descriptive? Maybe it's not perfect, but it's good enough for my purposes, right? And so there's something empowering about that when it's not a business yet. And also if the whole thesis is that AI is gonna change how we, how startups are built and funded, then how am I not gonna get my hands dirty, right?
So I wrestle with that just on, on principle. But my natural state is not getting my hands dirty. So it's like I'm wrestling with that. So having this conversation and having a blank, the equivalent of a blankie a PM who's closer to the tools than me, who might not be an engineer, but still closer to the tools than me, to help me through it, right?
That is that is better than me banging my head against the wall. Myself. But wouldn't it be even better if there was like a, a college hacker who's immersed in the tools, who can sit with me and get immersed in my processes and then help me connect the dots on, manage to the index of what we can automate as we go and automate more and more over time and have it be neatly organized and structured and high quality and performant without bugs and it's like that's where I want to get to.
And so it's like, all right, is that really going to be me that gets it there or? Am I just, am I just doing this as an interim step to get dirty enough that I get so frustrated to justify taking out my wallet, but I don't know. What do you think?
Paul Heayn: Yeah, and I think that's that's the fun of not having any like business plan or like revenue tied to it because you, it's your money out of pocket and you're just like do I want to spend a hundred dollars on this or whatever it is? Or do I want to spend the two to three hours to try to figure out how I can do it on my own?
Like transcription, like you could just use the. Or you can try to build it yourself and then it's like that's the other balance of do you want to, if you're building, if you're going to build this for something that's like self hosted, does it need to be on the cloud? Does it need to have all that stuff?
Or is it okay to try to deal with building this, these products yourself or these tools yourself? Yeah. And having to deal with the maintenance, and I would say that's fun for a fun project, but like, When you start to scale, I don't think it'll function. Like that, at that point You need an engineer, who's okay, yeah I can help you rebuild this, because this is not great.
Jason Jacobs: My, my dream is that I can just go and be a content machine and build tons of relationships, have tons of discussions like this, learn a ton. And then in my wake, the infrastructure underneath, which is constantly getting better, smarter, more automated, higher quality, et cetera. And and then the two go hand in hand.
Our technical infrastructure expands as the content The infrastructure expands and the content feeds the technical because the bigger the content library, for example and the better we get at structuring the data, the more insights that can be gotten in the aggregate and per episode in terms of guest prep, in terms of what topics to cover next, in terms of in the aggregate what's the consensus, right?
As this growing body of knowledge grows. What are the key themes that are emerging, right? There's so much there. And that could inform our internal planning. But that also could be content in its own right. Like living, breathing content as the library grows. And that's awesome. And oh, by the way, Those tools that we built for ourselves, that we dog food essentially, could be valuable to other content shops as well.
And then maybe there's a business with a revenue stream. And then all of a sudden we have a budget that we can then feed back into growing faster, that doesn't come out of the savings that I need to put food on the table because, because I don't earn a living, right? So that's the dog.
Chasing its own tail around, and one way or another, it's gonna get addressed and I'm just nudging everything towards the middle. I'm getting dirtier with the tools, having more discussions like this exploring different paths and I don't know the answer, but one of the things I was hoping we could do today is talk a little bit about where I should start.
Resourcing aside, we can talk about resourcing. Separately, but even resourcing aside, just from an infrastructure standpoint what tools should I use, what foundation should I set, how I should go about it, because right now, it's I feel like my MacBook is like my kid's backpack, his backpack he always knows where everything is on there, but that thing is a freaking mess and That's how I feel about my Macbook.
It's I forget which things I've subscribed to, they're all using the same words yeah, I gotta go to so many different tools for so many different things, and it's did I use my Gmail, did I use my work account for the login, and it's just, and I can't get anything to work, right?
It's like such a mess it's, it, for me, it's a long way from Legoblocks, right? And and that's what I want. I want one site to go to that I can just Legoblock my way to whatever the heck I want. Do I need to build that? Do I need to wait for that to be there that someone else to build? But until it's there I need a team, there's no doubt about it.
Paul Heayn: contact me. Yeah, and I think, so I recommended N8n because I think it gives the flexibility plus the UI to build these things. But I also spent like multiple nights in the past couple days playing around with it.
And I felt like the error handling, like when you get, when you reach errors, it's so frustrating to deal with because the errors,
Jason Jacobs: played, after you sent it to me, I logged in, and I spent a good few hours on that thing. And now I have a sense of what the stated vision is of what I can do. And it's oh, I see why Paul sent that to me. That maps pretty well to what I'm trying to do. I cannot get that thing to work! Ah!
Paul Heayn: Yeah, so so yeah And so I think I need to spend a little more time because there's a lot of like nuances in there so maybe like I'll come back once I figure it all out and show you how to use it. But I think that might be like where you'd want to get to because I think there's like an enterprise version of that, which you can build products on top of it.
So you basically use that as like a middle layer to whatever, services you're connecting to and then build like a website on top of that, that connects to it. So I think like longterm, that might be like something to invest more time into. But right now it's I wouldn't be able to explain it to you, but I think that is the right path.
And I think that there's also make. com, which is another similar one. I haven't tried that out as all at all. But some of the things that you were asking about, some of them are just like, can you use Zapier? Or like some of these like triggering kind of software to start the automation process and get this done.
And maybe, like I
Jason Jacobs: But each one of these tools has its own learning curve, different UI different looking, it's like, it's hard enough for me to get used to any one of them, let alone how to start putting them together. It I don't know, it's a, maybe maybe I'm just not that smart, but it is it is frustrating and definitely out of my comfort zone to try to navigate these different tools.
And I'm sticking with it because, it was frustrating once upon a time for me to learn how to, Ice skate, and then I played hockey for decades, right? It it was frustrating one time for me to learn how to ride a bike, or to snowboard, or to, or to do a lot of things, right?
And then, to start companies, right? And it's now, or host a podcast. Super nervous my first podcast. Now, hundreds of episodes later I might still suck, but I don't get nervous anymore. It's like muscle memory. So maybe this is the same and I just need to invest the time and not get frustrated.
So I'm not giving up, but at the same time, I'm trying to be intellectually honest about What my superpowers are and where I need to complement my abilities such that I can, go build a big important company as efficiently as possible and not waste cycles or shave years off my life in the process in areas that could otherwise be avoided.
Paul Heayn: Yeah, and I think I think there's two other tools that you should look into, and the reason why I mention that is because they go Hand in hand or they're the counterpart to N8n. So one is in, in Microsoft Visual Studio code, you can add plugins. So I don't know if you have code installed but you can add plugins and
Jason Jacobs: I installed more things in the last few weeks than in my entire life before that. I don't know if I have that one, but I have a lot of things installed.
Paul Heayn: And have you have you been able to use them?
Jason Jacobs: I use chat GPT constantly, all day long, so that's one thing I'm using. Every once in a while I'll switch to Claude if it feels like it's a little more of a Claude type question.
Perplexity I, so even like Deep, people say deep research, and I can never remember is deep research the perplexity thing, or that? And then, certain things are then baked into other things, right? So it's oh, you can access it there, but you can also access it over there. It's like, how do I remember?
I might be subscribing to multiple things that I don't need to subscribe to, because I already get the things I'm subscribing to over here, over there, right? It's, but for example, for guest prep, I think that the manual process should be to figure out which tools to use. Do I want to use Perplexity first and then plug it into JIT and then manually and then figure out how to automate that next and then build an agent, right?
But I'm still working through those processes right now. I know, for example, that I can go to ChatGPT and that's pretty good for most of my stuff. Descript I'm spending a lot of time in doing a bunch of the editing myself and that I've started to get down. I'm still learning, but I'm already functional and that's good.
And then, I've played with a bunch of this other stuff Zapier, or Repl. it, I had a journey with Repl. it, and what are some other ones that I played Cursor, I, I took a stab at. Devon, I took a stab at, but anything where it plops me into Terminal or anything like, that feels like Terminal, and I'm supposed to function from the outset with no context or guidance forget it, I'm dead in the water. Yeah.
Paul Heayn: How do I get started? Like, how do I get this thing to show whatever I coded? How do I get that to show? And like Python, it's very easy to like, understand what's happening. Like in PI and like the terminal, you just type in Python and then whatever script you want it to run and it'll run. Whereas like some of the others, like you have to compile it or like you have to run a, actual server to get it to run. And I think that if you want to go down the code path, stick to Python, like just, I'm just going to say I'm sure people
Jason Jacobs: So then should I put this whole journey off for the next month or two months or three months or six months and just go learn Python and then come back around? I'm 48, right? I haven't learned it yet this is where I wrestle, right? It's it's easy to say, oh, just go learn Python, but it's that is so far out of, like, where my natural, my, my natural state is when I wake up every day.
Paul Heayn: I think with AI, you might not have to learn Python. You might only need to learn how to interact with Python, if that makes sense. Like you might just need to understand the file structure. How like the methods are created and how to start and how to stop the actual scripts and like You might be able to get far
Jason Jacobs: Okay, so let me give you, let me give you three paths. One path is Jason and all his AI best friends chat GPT and whatever tools, right? And like committed, solo, no help, no questions, like muscle through, just sit in front of the computer and muscle through every day until you figure it out, right?
That's one path. Second path is oh, I'll give you four paths actually. Second path is do that for a while and then once you understand what you're banging your head up against, then try to bring in help with with some, with a better sense of where you need it because you've been taking it on yourself.
And we can debate about how much mastery I should have before it's okay to pass it off. And, like when to wave the white flag. The third path is wave the white flag now and just go get some help, right? And I guess, The third path would be get some help, and that help looks like a junior hacker, call it an intern, call it like, someone part time, but someone to actually sit with me and do the stuff and maybe sit with me remote, but talk with me, understand my processes, and then work with me to automate, which like, sounds awesome, to be honest and the fourth path is, no, still do it yourself, but do it pair programming style, where someone like you, or maybe someone like Big Philly C or, somebody we can talk about what the right background is, but that can actually do it with me and hold my hand and reassure me that it's going to be okay as I'm muscling through it so that any time I start to get frustrated, they can save the day.
Paul Heayn: So
Jason Jacobs: you do if you were me?
Paul Heayn: I would say I say path three but I would change it and I would change the Instead of saying I need an intern I would say, I rely on AI to ask the questions. You will get so many answers from the AI. And then augment it with phone calls like these. Or with social media questions about this.
Or like Stack
Jason Jacobs: you so instead of a human, you think that just leaning on the AI for help will get me
Paul Heayn: yes. Yeah. And that's like literally what I've been doing. That's in, in the
Jason Jacobs: We are not the same. Yeah.
Paul Heayn: but
Jason Jacobs: to be a web developer and now you're a product manager. I'm not either of those things.
Paul Heayn: But that was a long time ago, and I wasn't a great web developer that's fine. But the tools now in, in the actual plugins and the agents that are available are extremely good at helping or deciphering and reasoning what you're trying to get at. And if you get like the thought of a chain of thought that you understand what they're actually like dissecting and you can actually learn more from that.
And then they give the answer and you just, you like learn very quick. So I think that there's, it's a, it's like an augmentation, it's like AI with someone who could call. And maybe that's you, Pay someone hourly or something like that a hacker or something like that, like
Jason Jacobs: if no one builds that service, I'm going to build the service. It's like cameo for pair programming. You know what I mean? It's like a jukebox. You,
Paul Heayn: Like a stack overflow pay, pay
Jason Jacobs: it's like a, it's like a genie. It's you rub your belly three times and a pair programmer appears that can help you untie the knots that, that you've gotten stuck with, you
Paul Heayn: I love it. I feel
Jason Jacobs: But with consistency! Who's plugged in over time and has all the history and knows where your code base is and stuff but just rub your belly and there they are. I'm gonna build it I think, like, programming genie, something.
Paul Heayn: I love it. I love it.
Jason Jacobs: And I want the same, I want the same thing with with, on the other side, cause I've, I told you I've been talking about building that show and starting with character development and stuff.
I want the same thing with like some of these big social media comedian, character developer they post a lot, but it's never themselves, they're always like, in character, right? And they build big followings I want someone like that to call to help me figure out that too, but that's a, that's another story. I also want parapro yeah. Or even without avatars, just I don't know how to be funny. I don't know how to build skits. I don't know how to how to do a lot of that stuff. And people know, and also, there's increasingly tools that can help, but I need the same help over there that I need over here, right?
I feel like that's going to be a lot of areas, right? If you're a great lawyer, But you are like me as it relates to the tooling, right? Like, how do you start to automate the mundane stuff? You can't do it yourself. And also, you're booked on the treadmill of billable hours. How do you get there?
Same thing with accounting. Same thing with a lot of these things, right? And I don't, yeah, there's these services that are going to merge that are going to do it all. But I also think there's hybrids where you want to just start picking stuff off in house and make yourself more efficient.
But like, how do you actually get there? It's like a big, um, Business transformation question.
Paul Heayn: I think that People will, I don't know how like much traction still get, but I do think that people will start building personal software that is custom to them. And so I built like a workout app for myself. Like I, yeah, I asked chat GBT or I think I asked like perplexity to build me like a workout plan that was like explosive and strength training.
And I was like, I want to work out six days a week. And he gave me, gave it, spit it out and JSON. And then I was like, all right, can you build me a web app around that? And it did
Jason Jacobs: Did you tie it to your bank account such that any time you don't work out six days a week it automatically withdraws and donates to charity?
Paul Heayn: Not yet. That's that's on version two. No, but like this idea of yeah, I want to accomplish something that like, there's no barriers, there are barriers, like there, there are things that are like, not solved or you need engineers, but a lot of those, the problems that we have as people.
That don't have a huge TAM are things that people might go and try to solve themselves. I think that there is something there where you're, where it's yeah, I just need to get started, or I need to get help and have someone holding my hand. There is something there because I think people might be starting to see the light, but they can't get past the, like, how do I just get started, or how do I get past this error, or whatever it is.
Jason Jacobs: I'd like to go in two directions. One, I'd like to go big picture for a sec and just talk generically about the future. And then I'd like to come back around and talk about my needs, very specifically. Because I thought this was going to be a pair of programming show and it's a therapy show instead, but that's fine.
But on, on the big picture one I had a guest come on. Sahil Lavangia came on recently. And what he was saying was that that essentially vendor sold software is screwed because these tools are going to enable more and more companies to just build personalized stuff just for them, in house.
What do you think about that? Agree or disagree?
Paul Heayn: I agree, but I think that it's going to take a while just from being like in the corporate structure, like the contracts take a long time, they take a lot of like diligence and is someone in leadership willing, in a corporate level, willing to take the risk to, build it themselves Knowing how long things typically build in their company, like just, it just goes a little bit slower and will be competitive enough to do everything that the SAS companies that are already providing, are they able to build it?
And are they willing to put their name on the line for that? So I think that's going to take longer than we think, but I think that there's definitely the words are written down on paper and it'll happen. I think it'll happen in, I don't know, five, five years ish.
Jason Jacobs: And when you look at startups, and you look at bringing new products to market, and you look at the roles of engineers, PMs, designers, and the process in general, and granted, There's no fixed process. The process is different at different startups and some have wildly different processes, but there's certainly themes, right?
Directionally, how do you think AI is and will be changing the roles in terms of how many, the types of roles and just for you as a PM, like, how much is your job going to be changing directionally and how much has it already?
Paul Heayn: I think that with the economy and everything that's happening, people will be forced to use it to AI to make up the difference in lower head counts. I think that's just going to happen. So it's going to be a force that we have to deal with. And I think that the people who know how to utilize AI in an appropriate way. We'll be able to get answers to their problems quicker. When I say that, get prototypes in front of users faster, understand that's really the true solution. Prototype a lot of different things very quickly. And then say yes, we needed to productize that. And I think the, I don't think the triad of UX product engineering is going to go away, but I think that the roles are going to expand to overlap each other. And so I think that maybe you have a team that was typically, release it. We're focused on one thing. Maybe they're focused on three or four different things and like they're accomplishing what three or four teams can actually do.
I think that's what's going to happen. And it's all the stuff that, yeah. PMs, all the artifacts and documents that we have to create, I think will be created much, much quicker. And then you're more pro like proofreading what's actually happening. And figuring out how to get in front of people and that will be the bottleneck.
It's how do you actually test this? How do you actually talk with the users and like, how do you speed that up? But I think, yeah, like I think there'll be some like, need to embrace AI just with how things are potentially changing.
Jason Jacobs: And where does that come from? Is that just each person individually just tinkering and trying different stuff? Is there certain training that should be done? Like the training for how to be a PM. Should that training change? And should it be the PM training that delivers the how to incorporate the AI tools?
Or does there need to be AI specific training?
Paul Heayn: I think I don't know. It depends. I think that there probably needs to be AI specific training just literally from the past like week I was like, oh my gosh, like this is like Ridiculously like complex and trying to get this up and running So do I think that the tooling is there for anyone from any kind of?
Walk of life to pick this up and automate things away. No, I don't think so I think it needs more training but as for like specific PM training, I think PMs are over, and I can talk just about PMs, but like they're overworked, and they're looking for ways to try to be more efficient with their time.
So they're naturally going to start to entertain these things. And I imagine most PMs are already using AI in some capacity to try to do things. Maybe some other PMs have tried and like they didn't see the light and it's okay I'm just good. This doesn't fit my needs but like maybe they haven't dug deep enough And so I think as we go into the future Though there will be a need to Try to level up quickly and try to understand what these tools do and how do they fit into your everyday life
Jason Jacobs: okay, and to get real tactical here, so you said, me and banging my head against the wall with the tools and that's fine but let's take that, that N, is it N8N tool that you sent me to, yeah, so Sh Should I start with a tool like that or should I start with chat? GPT?
And if I should start with chat, GPT, how do I know if it's chat? GPT or Claude or one of the other a hundred? LLMs like we, when I wanna do anything, where do I start?
Paul Heayn: I think start with the one that you like and I know that's like a cop out
Jason Jacobs: The l Are you talking about the LLM or
Paul Heayn: yeah, LLM like just start with
Jason Jacobs: Always start
Paul Heayn: For for now. Yeah for now start there and I would say
Jason Jacobs: Uhhuh,
Paul Heayn: As someone who like doesn't want to spend monthly subscriptions figure out a way to move from chat GBT To using the chat G-B-T-A-P-I key because one like the API key, you buy credits and like they, they don't expire, at least all the ones that I've used.
So build your own chat, GBT in a web browser that you can use and you don't have to worry about a monthly subscription as much. And that way it's like a fixed cost. And then I would say learn about model context protocols, which are what agents use. They're the tools that things are used that agents are basically talking through APIs.
So Claude built this protocol and there are a ton of different. Websites that you can integrate with this MCP. And so the agent that I use for coding, it's called RooCode R O Code. And you can turn on MCP servers, and basically give it access to a lot of different things, and all of a sudden your questions not only are what's in the LLM, but it answers with what's in your environment.
What's on your terminal and also it'll search online and pull out documents or documentation from like GitHub repos, or it'll give you like, number of different options and like pros and cons there. So I can show you like the tooling that I have, but I think that like understanding like, MCP and then understanding How does this all work?
What does this do? And having the visual reference, that's what like N8n does, where it gives you the canvas and like it shows you the flow of how things work. But then understanding okay like I can just add tools that are, built into N8n. But then the power comes from the extendability.
Where you can, one, either connect different workflows together in NAN, or you can build your own tools that like extend even further than what you're doing. And so like that means like you can build your own Python to like search for other podcast episodes and, have another one that like transcribes it on your own, on its own.
Yeah. Rather than using like ChatGPT, again, like I don't use ChatGPT. I have Gemini right now. So that's why I'm like trying to figure out like how to go about these different paths.
Jason Jacobs: like for me, if I just look at my podcast production process as an example, because that's the main thing I'm doing now, there's just a lot of moving stuff around. It's like I save it here and then I move the files. To Google Drive, and then I upload them from Google Drive to the script, and then I, add them both to a single project that gets created, and then I there's a whole set of steps, and I sat down and mapped it out, but I haven't sat down and really mapped it out but if I really mapped it out, I have all the steps, and then it's alright wouldn't it be great if just, When I hit stop in Riverside, it just went and did as much of that as it could, and it's just a hundred less things that I need to do, right?
Or, and even if to start it was three or four less things I needed to do, and then over time it was five, eight, ten, twelve, fifteen, and then just started gobbling up steps, it's then my life's gonna get easier and easier, and I'll be able to cover more and more ground, right?
Perfect.
Paul Heayn: and I was I was looking into Riverside Descript Riverside doesn't see so Riverside doesn't seem like it has that many APIs that you could use, I think it was Riverside. And going back to the MCP kind of thing the model context protocol, that's just basically a way for it to hit endpoints on an API.
It's just hey go to this Google API or go to this air table API and do this. That's all it is. And the LLM is actually figuring out the reasoning on which tools should actually use.
Jason Jacobs: That's another thing. Each step, it like lays out 15, ChatGP2 lays out 15 steps I need to do, and then each step, it gives me 7 choices of tools to use. And it's do the math on that! It's 15 steps and each one has 7 choices to use. How many tools is that I need to understand to know? What makes the most sense.
So I can grab the one at the top of the list, top of the list, top of the list, top of the list. But then it's oh, we left out that those two aren't compatible. Then there's like all these gotchas that you find, right? So it's just, I don't know. Maybe it can ultimately get me there.
But I find that every single step, it's like, screenshot. Dump the error into the thing. Screenshot, dump the error into the thing. It's oh, the instruction says I'm supposed to do this next thing, but actually, there is no choice for that on the screen I'm on right now. Screenshot! And it's oh, I see, and then it, yeah, it's just, it's to try to move 50 feet, it takes me 10, 000 steps.
It's, at least, where I am now. Maybe that's just because I'm still in the beginner mode or whatever.
Paul Heayn: Yeah. And I think because you're in chat, GBT, it doesn't see what's in your computer or what you're looking at. And so I think there's two angles. There's one, like the coding and like allowing it access to your directory or allow it to ask you for access to write to your directories or get things from your directories.
And then there's the browser use, which is like an agent that will open up browsers and Look online and actually click and walk through different paths and get to where you need to go. And so that is, I think, where some of these tools where you can't have API access, it's where you're gonna have to probably use browser use, or it's called computer use agents.
Browser use is just the repo that I had to use. The only issue with Browser use or I've looked too much into the other ones, into other solutions is if you want to change things you now have to go back into code. And so that's now I'm dealing with someone else's code.
Do I want to do that? Yes or no. Like I'm not a developer anymore. I'm not, I wasn't a developer before. Like I wasn't like a a Python developer before. And so that is one of those things where it's yes, like you could solve the problem, but it'll probably take two or three hours rather than a couple minutes.
And is that enough time or is that like an appropriate amount of time to automate it? And then slowly you get away from using the computer use agent and try to hard code it. But I think the computer use might be How you can get around some of these, like I'm clicking on everything and I have to go to all these different places.
You might be able to automate some of that with that, but then to your point, you're adding another tool onto your toolbox that you have to remember and run and I don't know if it's like the perfect solution. So I think I'm going back to N8n as like where you want to get to. But you also want to learn Python.
Uh, in the short term.
Jason Jacobs: Okay,
Paul Heayn: So
Jason Jacobs: go ahead.
Paul Heayn: do you want me to show you the coding tool that I was using?
Jason Jacobs: Sure. Yeah, do you see anything on your screen that like at the bottom on the right side does it say share?
Paul Heayn: Yep.
Jason Jacobs: Yeah, so you should be able to share your screen.
Paul Heayn: All right. Let me see if that works. Window. All right. So I am just going to create,
Jason Jacobs: Yeah, and we're going to publish this as audio and as video, so for the audio only they won't see this, but for the YouTube they will.
Paul Heayn: so just creating this guy and let's say we want, okay, so first of all. This is VS Code Jason, you've seen this, right? I don't wanna
Jason Jacobs: Yeah.
Paul Heayn: Okay so I am gonna open the terminal to know what's going on. And then I have Roo code installed, which is down here, this guy. And so I have this on the left hand side.
Or the right hand side. And so then I, you have a couple different options from an architect to an ask. So let's just say we want to build, um, Build like a Engineering connection social networking site. I don't know, build a Social network. Oh. So hold on, let me cancel this because what I wanted to do first of all was actually do this in architect mode.
Jason Jacobs: So is Roo like Repl. it?
Paul Heayn: I don't, I thought Repl. it was like an online testing tool.
Jason Jacobs: Repl. it is like natural language to code.
Paul Heayn: Okay maybe? So I'm gonna say keep it simple. Only build so I don't think it, it would know what we're actually doing, but that's okay. This is more just like doing the steps of asking the architect and then saying
and I have all the, like the auto proofs on, maybe
so I don't know why. So is this like similar to what Repli does? Like it'll write the files on your computer and,
Jason Jacobs: Yeah, it looks similar to what Repl. it does.
Paul Heayn: Okay then.
Jason Jacobs: to my untrained eye.
Paul Heayn: All right, then we don't have to go into too much detail then. I thought it was pretty awesome to just basically,
Jason Jacobs: it is awesome, but then trying to get it from that, it's oh my Repl. it experience was in ten minutes, I had, it looked like the thing was built it built my website, and it was like, oh, it has all the tabs, it has everything it it even took a stab at using the copy putting my desired copy into its own words, and it did a better job than me oh, this is amazing, and then trying to get that thing into production was the biggest nightmare.
Paul Heayn: you want to pull it up?
Jason Jacobs: I don't think I still have it. I think I killed it. Yeah. No, because I just went back to Squarespace because Squarespace was just like Lego blocks, right? And like Lego blocks is where I live.
Paul Heayn: got it. So maybe coding isn't the right path. Maybe it's, maybe it is just N8N and, utilizing chatGBT to help you. When you get these errors or like to learn these kinds of things or, um, get through pain points you like call on people that you know, or ask those social media.
There's gotta be like any and like groups that you can join.
Jason Jacobs: I might, maybe for starters I can just I can just write down all the steps to my process, like maybe that's a good starting point and actually try to document it accurately so that I can hand it to a human or to a machine and say here's what I want to automate, like what aspects of it can be automated today.
Paul Heayn: yeah, and I think once I wrap my head around the differences between all the different agent types. Yeah. I think I can get, make some headway on it. It's just the error handling and what's actually happening between the flows of data is what's throwing me off. But yeah, once I get that I'll let you know.
I'll tell you what I did wrong.
Jason Jacobs: Cool. This is awesome, Paul. I'll report back to you as I get further along and definitely report back to me as you get further in your experimentation as well, both your kind of hobbyist experimentation and the way you're incorporating it into your job. But it's really fascinating to hear how much value you're already getting from it.
And I can see a path where I'm also going to get that value, too. In fact, I'm very confident that I will. It's just a question of how that manifests and what order I do things. And that's what I'm in the process of sorting through.
Paul Heayn: Yeah, that's awesome. Yeah, thanks for taking the time. Appreciate it.
Jason Jacobs: Yeah before we go, anything I didn't ask that you wish I did? Or any parting words for listeners? Anyone you want to hear from?
Paul Heayn: No, I don't think so. I I'm interested in learning about. up and coming AI companies. I thought it was super interesting when you interviewed Lovelace, because like I was like, I had never heard of them. So I think that's super interesting on in general, just hearing what people are doing with AI.
Jason Jacobs: Great! Then, stay tuned to the pod because I'm gonna have a lot more on like that.
Paul Heayn: Awesome.
Jason Jacobs: Thank you for tuning into the next, next, if you enjoyed it, you can subscribe from your favorite podcast player. In addition to the podcast, which typically publishes weekly, there's also a weekly newsletter on sub stack at the next, next dot sub stack. com. That's essentially for weekly accountability of the ground.
I'm covering areas I'm tackling next and where I could use some help as well. And it's a great area to foster discussion and dialogue around the topics that we cover on the show. Thanks for tuning in. See you next week!