Roman Gun AI === [00:00:00] People's core competency is gonna be orchestration and managing a combination of humans and agents. Those agents, they're going to need a set of soft skills that maybe a lot of the best engineers today don't have. As these systems evolve, soft skills that may not have been as valued in a lot of tech companies are going to get valued more, and you're going to have to bring in a new generation of people that both have technical chops and soft skills. Welcome back to Launch Pod. We're kicking off a new series on how actual product leaders and their teams are using AI to work smarter and faster. I've been meeting with product leaders across the country and the same question keeps coming up. How are others actually using AI at work? This series is all about those answers, no hype, just practical workflows and tools your peers are using right now to level up. In our first episode, we're bringing back Roman Gun VP of Product at Zeta Global to talk about how Roman uses prompting and AI orchestration to automate writing PRDs and turn them into truly collaborative tools, the [00:01:00] communities and platforms he uses like Reddit, discord and beyond. To learn about cutting edge AI tips, tools and early access, and how his magic the Gathering side project actually turned into a proving ground for AI experimentation. So here's our episode with Roman Gun. Jeff: All right, Roman. What's up man? Good to have you back. Thanks for coming back on the show with us. Roman: Dude, it's a pleasure. Had a blast last time. I'm super excited to chat again. Jeff: Yeah, I'm stoked 'cause we're, we're doing something new this time. We are adding a little bit of a sub show to our show and we are going to be talking about real world, real life applications of ai. Not for your end product, but that people like you are using with your teams to move faster and be, stronger and more efficient and smarter and all those great things. But, these are the things that hopefully people can take away and be able to use some of the stuff we talked about to actually make their teams faster and better. I've, been traveling the country talking to product leaders in cities all over and having dinner with a bunch of folks, and this has been the number one conversation that has come up [00:02:00] is \ What is your team doing with ai? And it just turns into hours long conversations with people across these tables. So we wanna do it here. And, and the first one I thought of was you, Roman. So, stoked have you on and, and let's get moving. So Roman: DAF punk this. Let's go harder, Jeff: yeah. Roman: stronger. Jeff: I, I, I'm glad you caught that reference. We talked a little bit before this just to be like, this is the idea we have, let's flesh it out together. And about 10 seconds in, I slacked our producer m on the side and was just like, yeah, okay, we're good. This is, this is the first person we wanted do, 'cause you brought up, you're like, well you're on the notebook, LM Discord surfer. Right. And immediately, had exploded. 'cause no, as advanced as I thought I was with, using these things and, and working on it. I think there's this first piece of just like, community is a huge part of AI and, and how you learn and how you kind of see what's going on. So can you maybe, can we talk a little bit about that to start? Roman: Absolutely. So I think there's two primary channels for being a part of the AI community that I use. Jeff: Mm-hmm. Roman: those two, particularly being Reddit and Discord. Jeff: [00:03:00] Yeah. Roman: so Reddit is great for a bunch of articles, studies, use cases, PMs. Complaining or showing their victories. It, it's a lot of fun because you get to see different takes on it, right? Like there's subreddits that are dedicated for people who, think a GI is gonna be here in 12 months. There's one for, it's gonna be here in five years. There's ones for, oh, it's not gonna hit until 2035. That's the only one I really disagree with. , it's really interesting because you get to. Choose your different level of zoom, whether that's on a product level, 'cause every AI product has it or a specific sub theme. And by the way giving away my secret here, but one of the great way to get free trials for all these, for like a year is to be on those Reddits. They constantly give us subreddits. They constantly give out. Codes so that you can just get in there for free. And then you can go into like the even deeper community on Discords, and that's when you get invited into like the beta programs and you get to start using all these new tools before anyone else. And I [00:04:00] think that's really cool. Remember when I was a younger man used to have like a different version of either. Any meta product, whether it's Instagram or Facebook or paper, still one of the best products ever that that failed in my opinion or LinkedIn or whatever have you. And I think a lot of that differentiation went away, either because platforms matured or because I just outgrew that segment. And now with this AI revolution being a part of these newer communities testing these things, it, it's been a blast. Jeff: Well, it's fun 'cause like it gives you the ability to kind of, the people who wanna be a little bit more bleeding edge. I think I jumped onto the, the notebook LM Discord, which by the way, notebook lm one of my favorite tools to use. We use it actually basically every time I do a podcast episode is key to, making this show work. But I jumped on the Discord legit under five minutes in it, and I had already discovered that they were coming out with a mobile app. And there was a early access, beta tester for the mobile app. And I missed that by like. [00:05:00] Hair, but I'm stoked for that because I love that thing. Roman: sure my test flight has never been more active. So that's fun. Jeff: But this is a, staying ahead on this stuff is, I think, really, really important. And we'll talk about why. But we do these kind of like product leader dinners around the country and a guy named Ryan Johnson, who's been on the show before, he's a CPO at CallRail, was talking through how his team uses ai, but he's talking about, he has PMs, shipping code, he has SEOs that are automatically producing and launching pages. And, and the point was just kind of, if you're not learning how to do this stuff and, and making yourself move faster and learning and staying ahead. You are going to fall behind and some of those gaps are going to be impossible to make up at some point because the people who just keep learning and get ahead, like they're gonna keep going. Roman: For sure. And I think to that point I do think that's gonna evolve from doing more to orchestrating more, Jeff: Yeah. Roman: I, I have a strong belief that people are going to all become managers. Whether that's of, beings or synthetic beings, that's [00:06:00] probably gonna be a mixed bag, Jeff: No. think that is absolutely where it's going. Everyone's gonna have to orchestrate an almost an entire, like, business unit of their own. It's gonna be fascinating. And I think we have a great example we can just kick off on, right? Like, I have never met a PM who enjoys writing PRDs. And every time I've ever brought it up, people laugh and cringe and, and all those things. But this is one area you've already. Made a ton of progress on 'cause like every normal sane product person on earth, you don't like writing PRDs. Roman: If you find a product person that likes it, run, there's something not, Jeff: They definitely have a hand in their freezer if they like, if they liked it so. Roman: Bundy vibes there. Just don't do it. Jeff: So how'd you get out of it? Like, walk us through this process how's it work and, and what are the unlocks that you found that really made it possible? Because I think, there's a lot of people who say like, oh, it can't do that. It can't write that well, it sounds like trash. And, and I think there's a lot of ways you can make this stuff work Well. Roman: . Well, I, I think a lot of it is just going back to the core [00:07:00] of it, like what is the purpose of an Epic or a one pager or a PRD? And at the core of it, it's communication. It's a really a human soft skill, Jeff: Mm-hmm. Roman: I. I find that teams who are really on the artifact of, I have this PRD or I have this epic, or I have this one pager, do it for one of primarily two reasons. One they want a security blanket. Jeff: Mm-hmm. Roman: to the process having been done that way. or two, they just wanna check a box, right? And, in some organizations that works, but to me that feels very waterfall. So having all this documentation. Before you collaborate and, and really express the why and how we're gonna get there always felt a, a little disingenuous to the Jeff: Mm-hmm. Roman: So what I always did is I had conversations with people and then we would build these things dynamically based on what we decided is going to work and not going to work. So this thing doesn't have to get [00:08:00] revised 13,000 times, but like, 9,000 times. that's where actually the basis of this is with how I build PRD. So I have an agent for epics one pager, PRDs, weekly updates, what have you, all that good stuff. And essentially I take examples of PRDs or epics that are really good or in the format that a company likes. As well as that I believe makes sense, Jeff: Mm-hmm. Roman: fed those examples. Also just. From there I indicate what I want to get across in a Jeff: Mm-hmm. Roman: A PRD. And the way I express that via the agent for myself is A-T-L-D-R. Jeff: Mm-hmm. Roman: of the time when you're speaking with a higher level, person, you wanna just go and dive in and say, Hey, what are we really achieving here? Jeff: Mm-hmm. Roman: So I start there even when I'm uh, doing this. Jeff: Right. Roman: never actually type this. I'm always communicating back and forth. So now the agent knows what kind of content it expects from me, and it can indicate right back to me, Hey, I don't have this piece of [00:09:00] information, and then I can go ahead and provide that. So now I'm having a conversation the same way I would be having with potentially an engineer or a QA manager or a designer. And by the way, anytime you do this, you do still have to specify. This is for design, this is for qa, this is for engineering. And. Back an engineer and a front engineer might also want these things differently. So it's right back to communication. Once you do that, you make sure that it's outputted in the same format. So if you're using say, confluence or Jira and things come in a table form or broken down into specific sections, already have that formatting so you can copy and paste it where the people are going to be reading it. Advanced points for using an action to. that right to your specific instance. So you can automate that directly in. And then from there you communicate with people the good old fashioned way. Say, Hey, this is what I have, this is what we're trying to achieve. Do we have any questions? And from there you have that transcript running while you're having this conversation. [00:10:00] See that right back into it and say, Hey , these are the conversations we had. How do we update this based on what we discussed? And now you have a living, breathing document that's constantly being iterated on. That to me is how you evolve. The PRD. Jeff: Yeah. Roman: it's a good artifact to checklist against, but it needs to be built organically and we're building some synthetic and organic synergy here. Jeff: I mean, I do think that's an important part. It's like we didn't start all doing PRDs for no reason. They serve a purpose. You want them done well. You want the right information so people can all be on the same page about what we're building, why are we building it, what are great outcomes, et cetera. But like, I, I think there's a couple pieces in here about that you went through that I just wanna kind of dive in on and clarify how you're doing. It's like a, are you actually using the, the voice interface? Do you mean by, you're saying you're not typing, or is it just more like you'll have the conversation, record it with, peers and you'll upload the transcript or the recording or like, what does that actually look like? Roman: Step one is I use the advanced uh, voice functionality Jeff: [00:11:00] Yeah, it's so Roman: and I can just keep going and it synthesizes it. It's wonderful. So step one is that then when you have that baseline, you then go and have say a team meeting uh, that you record and you have a transcript for. And then after that, you feed that transcript into that same agent where you had that conversation where you created that PRD and you say, Hey, based on these, what are the main takeaways? What are the things that need to be updated, revised, et cetera. And that's when you do go to text when you're doing the copy and paste. so then it spits that out, and then you can go into right into conversational mode with it right after that to say, okay, this is how we should tweak it. This is something that we can fade out, et cetera. So you go from talking to the ai, talking to people, talking to the AI again, and then you ship that right back to the people. It, it's a Jeff: Yeah. Roman: of human synthetic, human synthetic, human, synthetic. Jeff: We're all becoming just cyborgs. Uh, um, Roman: The transhumanism thing. Let's do it. Jeff: You talked about a little bit earlier in the process, you fed in what you wanted, right? Like this is examples of great output. Maybe this is examples of bad [00:12:00] output. I don't know about you, but I, I found in that instance it's training A GPT to kind of write the way I want it to is almost at times like working with a 6-year-old or something, where early on it's a lot of. You put in what you want, you put in what you don't want. You kinda label that, and then you do some test outputs and you kinda have to go back and go like, no, no, no, not like that. I said this. Here's what was good, here's what was bad. And you just go through that cycle like 20, 30 times. The more you can do it, the more it dials in and just each time be like, that was really good. This was bad. Don't do that again. Roman: My GPTs and my three-year-old, very similar in that way. It's Jeff: Yeah. Roman: like constant loop, constant reinforcement and then cycling through it. So a hundred percent. Jeff: And then once in a while it forgets and you have to be like, no, no, no. Remember we don't do that. Roman: The, the AI doesn't give you a cheeky smile, though. The, the three-year-old will be like, oh, I actually knew I was just testing my boundaries. Jeff: So through this, right? You train, you train the GPT on what you want, what you, what's, what's not, right? And [00:13:00] then you kinda have this conversation, but it is this kind of cyclical thing. But you're able to build this now living, breathing. Output of A PRD that updates live and, and, right. You talked about some integrations with, JIRA. I think that's what you said you guys use. But this is, this is something almost anyone could do. It just takes a little bit of time and focus, right? Roman: Again, a little bit of time, but it saves you so much time Jeff: Yeah, Roman: I think a lot of these tasks are, hey, in invest, I, I don't even wanna say a, a day, Jeff: no. Roman: maybe a day. And all of a sudden you're saving hours and hours and hours every week. It's very much worthwhile. And when you do that, you actually learn things like, maybe when I did this PRD, gPT had actually helped me then build a great bartender, GPT, which we actually have for a company event. We had a station where you would talk to it and figure out the best cocktail to make to you, Jeff: Oh, that's awesome. Roman: you would go over to the bartender and be like, Hey, the AI recommended, this is what I drink right now based on my mood. Jeff: Right. You might take a day to do this, to [00:14:00] train it. I think that's true. If this is one of the first things you're doing with AI and kind of trying to use it in this way, it, it could take you a day. But I, I read a post recently talking about like the, I forget what five steps to, to get started with ai if you haven't or something. 50, I dunno. Some number of steps. But the first one was. I loved this bit. The first one was just start using it because everything you do builds on itself and it's going to get easier, right? Like you could, the bartender one was a lot easier to do 'cause you've done the PRD one and there's a lot of these learnings and practices that you will pick up and, and carry forward. So like each time you're just gonna get faster and faster and faster and better and better and better. And it just, if you haven't done something yet and you want, and you're a pm, automate your PRDs. Let's not be perfect first, but you know, like you said, a a day of work is going to save you exponential time down the road. Roman: Again, harder, better, faster, stronger. Cue the deaf Jeff: Exactly. We talked a little bit about like, one thing that AI has also made possible more than ever before is this idea of like, it used to be maybe you could wire frame something. Maybe you could, I. Draw something, [00:15:00] way back in the day. But now with all these kind of tools coming out with, the cursors and the windsurf and lovable and blah, blah, blah, blah, you can connect 'em to Figma. You can do like high fidelity prototyping in a way that is actually, you can create based off the modules that your product actually has. Or you can, build something and show that it's possible and you can make a functional prototype at least. And, and that's I think, a really important distinction about like. This helps teams move faster and make it more real. Be again, better, faster, harder, stronger. So maybe let's take a sec here, but like, how are you doing this? Roman: I think there's various tools, but one of the things that's coming right back to that PRD, you have to understand what you're trying to achieve and all the various interface types or buttons and interactions that you're going to have. From there you go ahead and you can feed in your component library or just a screenshot of like basic modules that you have on like say a, a dashboard or a audience building tool or whatever , and essentially you say. Okay, this is what I'm achieving. Use these component libraries. Go ahead and [00:16:00] generate this image. It will generate this image. it does that, you can go ahead and put into Figma and use like Coda or something else, and it'll like actually just generate like a baseline code for that. It won't always align perfectly with your components. But it's, it's a baseline. But even when you're not. Shipping it into production, it gives you a real narrative tool. It gives you Jeff: Yeah. Roman: you can actually get feedback on. Something that you can test with users, that the users don't care, that it doesn't perfectly align with your component library. Right? It's like they're testing, they're understanding, they're trying to give you feedback, and this just truncates that time. And you know when, when you're interacting with design, sometimes it's great to be like, Hey. this is a baseline and then you can go and take that and level up on it. Right. And I, I think that's been really cool. Jeff: I think that's important is it helps make things real and a lot of times I think a lot of people struggle to really like picture something that's not real yet. And so making the case for why a new thing is going to work might just, it's not [00:17:00] that people are, are not smart, it's hard to kind of like envision something that does not exist. But if you can take that and kinda make that first a step, like you said, it doesn't have to be totally functional. You don't have to ship it to production or anything like that. But kind of taking this, describe what you want, make the image. From, from one of the AI image gens, throw it in a figma and then, zap that into code. Now you have a semi working prototype that makes it so much more real. People start to, I've literally seen like engineers. Eyes light up on topics that they have been really anathema against Roman: Yes. You just, you get to that starting point of the creative act quicker, Jeff: Yeah. Roman: That creative act is when you align the pm the designer, the engineer, and they start bouncing ideas off of each other. And actually, to your point, building that, that's the entire point of this. It's not necessarily. We're gonna replace all of this or all of that, because I don't believe that's what's going to happen. It's, Hey, let's get here quicker so that we can Jeff: Yeah, Roman: on the more meaningful part of our jobs. Jeff: right. It's, it clears a lot of the crap out of the [00:18:00] way that we used to have to do so we can have the meaningful conversations okay. There's one thing that we talked about when we were kind of chatting earlier that I wanna make sure we hit. It has nothing to do with what you're gonna do in product aside from probably. The learnings from doing a project like this are going to help you in your day-to-day workflow still. But you brought up magic the gathering. I, I played a ton of that, when I was younger and still have boxes of cards hiding somewhere at my parents' house. But you actually built a project that, that helps you do this better. Roman: Absolutely. So I play in a format called Commander, which is Jeff: Yeah. Roman: a fourplay multiplayer game where you're all just, free for alling and I. I only started playing this last November and when I would go to like a local game store, like these people have been playing for years or whatever, and I was just like, whoa, okay. These guys like know everything. So I was just like, Jeff: Yeah. Roman: how do I level myself up immediately? Jeff: Mm-hmm. Roman: I actually built an agent that would help me build my own decks [00:19:00] because I have this thing where I don't want to do what everyone else is doing. Jeff: Yeah. Roman: I wanna do things my own way and my own flavor with my own way to win or what, whatever have you. I. Essentially I, I built an agent that, or I built a series of agents to help me build these decks. And it started by creating rule sets like, Jeff: Mm-hmm. Roman: you could only, you have to have a hundred cards. They have to adhere to the commander's color, identity things of that nature. And what's interesting is more so than in most things, I actually found a lot of context window limitations here. Because if. Your exact deck list as it evolves, as you're making cuts, updates hard to keep all that in context. It's hard to adhere to all the rules. These various models don't actually know all the latest sets they were released. What was cut, not cut what was banned uh, I meant to say. And it becomes interesting because you actually start thinking, okay, how do I want ingest real time information into [00:20:00] feeding this tool new cards? Then how do I make sure it does everything legally? Then how do I make sure it doesn't just give me generic, these are the best cards, but like, what makes the most sense for what you're trying to really achieve? And it became really interesting because I would say, okay, I, I did this with open ai. Now lemme try to do this in Gemini. Let me try to do this with Claude. And you see such like. Different nuances. Like for instance, like. I, I found the best successful open AI here. For whatever reason, Gemini didn't perform as, as well here. But I found marginal differences when I was using like, deep research versus anything else. Jeff: Yeah. Roman: really in interesting use case where you go ahead and you're like, okay, I didn't think this would be something that this would stumble on. Uh, you know, You can build a wonderful code but you know, essentially. Building cards for what was the kids game that for whatever reason, adults play. You stumble on and that becomes funny 'cause you have to just start solving through all that. And it, [00:21:00] it's a lot of fun and you start understanding like, oh, if I have defeated these cards that it doesn't know what's the difference between doing a live video versus taking pictures. And if I do take pictures, you realize, oh, it actually is only good if I take pictures of six cards. If I take pictures of more than that. It loses that in context. And then you start like formalizing that when I'm just like, okay when I'm. At work and it's just like, oh, people wanna bring in this report for that data visualization and this and this. I'm like, well, okay, let's put like a hard six limit on this. Because when I was playing with my cards trying to de-stress, I realized like the image processing caps here in an efficient way. So let's bring that in. And again, it comes back to the point that you were making that. Every project that you take on, whether directly related to the next one or not, is going to have some form of learning. And the more you embed this into your various personal and professional work streams, the better you get. . There's just always something to learn and apply. Jeff: I think one of the big things I've found is break it down into discreet steps and do [00:22:00] this step and then pass it to another thing that does this step. It's still not great at do this, then this, then this, then this, then this, and look at this. So, Roman: Absolutely. You, you need a series of agents, so maybe I have a, a deck building agent. Then I have like a game analysis agent that Jeff: yeah. Roman: creates multiple. Both decks that fight each other using one of the decks that I had to like do an analysis on it. From there it's like an agent that will go ahead and critique all that and update the deck. So, yeah you can create workflows for all of these things. Jeff: Yeah. Roman: and I think that's where the age of Agentic AI really Jeff: Yeah, Roman: when people start realizing that they can operationalize most of their lives to get to the part of it that's the most fun for them, Jeff: yeah. Roman: focus as much on the. The processing and the Jeff: Yeah. Roman: Right. I think that a lot of people, because of limitations in our tooling, have become so on their process, Jeff: Yeah. Roman: they, without the process, they weren't able to get good outcomes. But the point [00:23:00] was never the process. The point was the outcome. So let's accelerate you to the outcome. Jeff: Yeah. Back to, something I've talked a lot about, on the show recently is we're not here to actually build software. Like that's not actually the job. We're here to solve customer problems. Software is the way to do it, but at heart, no one's gonna buy you a thing if it doesn't solve the problem. So that's, that's the core North Star goal is like, discover the problem and fix it and, fix the problem that's worth fixing. I gotta ask though, , are you actually simulating games? Like you have agents playing and simulating the decks That's sick. That's so cool. Roman: It does like a four person game where it goes ahead and it has, like, it makes sure that it's different types of decks so it knows, oh, this performs well versus the nigro deck versus a control deck versus like an aristocrats stack or whatever. For most of y'all, none of what I just said really matters, but, Jeff: Yeah, Roman: it Jeff: I got it Roman: just different Jeff: though. Roman: strategies and it's just like, okay, it's gonna do well against this bad, against this, good against this. And it just helped me get to place now where feel like I'm a [00:24:00] decent player. Jeff: that's what I was gonna ask. Are you competitive now? Roman: want some tournaments. Okay. Jeff: Nice. There you go. Roman: I want some pre-release events. , it helped me refine how good I, I was at identifying and playing quickly in order to build these agents, I had to understand the core of what makes something good or bad. I'm still not great at four deck building, but I'm pretty good at like. Getting new cards when everyone else has them putting 'em together and just winning. I wouldn't have been able to do that without learning through these tools. And I Jeff: Right. Roman: own tools to learn. And that's the really cool thing about this. There's no more barrier to entry to learn anything Jeff: Yeah. Roman: really want to learn it. There never Jeff: Right, Roman: It was just much harder. But now Jeff: right. Roman: so, so much easier. Jeff: The barrier to entry was just the effort you had to put in. And now that's come even before like these, lovable and stuff came out. I remember chat GBT helped me build a Chrome extension that I was trying to do some like esoteric little thing. No one to built it 'cause it was such a stupid, like, singular use case I wanted it for, but I just did it a lot. But it, it, I [00:25:00] whipped it out and was done and. 30 minutes and I had a working Chrome extension that I was able to use one thing that kind of threw me for loop is when we talked is you said bring up something that, that I hadn't talked to you about before. It hadn't prepped you on and, and was unexpected. , and now it's time for that part of the show where we're gonna , throw a curve ball at Roman and uh, just chat through it. But. Roman: to come from my training data. Right. I. Jeff: exactly. That's, I mean, this is how you get, this is how you build resilient systems, is throw unexpected things sometimes at it. But this is this one that's come up a lot recently. I've been really thinking about this. I, I am curious, I think you're in a good position. I know in a lot of cases I'm talking to people who are saying, either our hiring of junior people is way, way, way down. Or more often than not, it's, we're not hiring junior people anymore because , the AI tools have basically enabled us to offload a lot of that work that they would be doing to the AI agents. And we're being way more efficient and we're saving money and stuff 'cause we're not hiring junior people. There's all sorts of, ethical stuff we can talk about there that we don't have to. But my big question is like. What if we wake up in 10 years and find, we don't have directors, [00:26:00] because at some point we were all junior folks doing those simple, rote kind of tasks that were time consuming. But you have to do that to grow into mature leaders. Like some of those people are going to grow up to be, the directors and the VPs and the CPOs, but not if we don't let them start. So that's my question. Like how are you thinking about that? 'cause I'm, I'm a little bit worried. I'm gonna be honest. Roman: It's a really good question. Jeff: I don't have an answer here, by the way. Roman: So I, I think we have to think about this coming back to like, what's the root cause of this? Is some of this ai? Yes. but I think a lot of it is also maybe a correction for the COVID years where tech companies were over hiring. Jeff: Yeah. Roman: I also think we're in a place where our economy is volatile. So. Businesses are pulling back a little bit. So I, I think it's a mixture of factors. Do I think that there's going to be less core junior engineering hires? Do I think there's going to be less junior hires [00:27:00] overall? Not entirely. The reason why I think that is, I think that now. You have this really clear delineation of roles where I just do software, right? I just do hr. Right? And I think that in the future, these sorts of roles are actually gonna start converging because back to what I spoke about at the beginning of, of this chat. I think that we're gonna get to a place where people's core competency is gonna be orchestration and managing a combination of humans and agents. But you're probably gonna start by not managing humans, but managing agents. and those agents in some ways I. We're going to need their own hr, right? Jeff: Yeah. Roman: to need a set of soft skills that maybe a lot of the best engineers today don't have because it took a different type of person to get really far along in that field. I think that as these systems [00:28:00] evolve. Soft skills that traditionally may not have been as valued in a lot of tech companies are going to get valued more. And going to have to bring in a new generation of people that both have technical chops and soft skills. And I think that is going to happen. So you, I do think that there's going to be a hiring. Just think it's gonna be a new breed of hiring for maybe, hybrid jobs. Which, which I don't mean like a remote hybrid. I mean that you're gonna be focusing on a more lateral plane of execution, like an entire workflow and responsibility set versus a more vertical, linear path where you just have to be like a very good systems expert. So. I don't think we're gonna get rid of all the juniors. think we're gonna have to hire a lot of new types of juniors. In fact, I would argue that in the future, at some point maybe a batch of these hybrid juniors might be more valuable than really the Ihap people. But I think [00:29:00] we're still a, a far away from that. And I think we're getting a combination of both the T's and the i's, but we're gonna move towards the T-shaped society is what I think. Jeff: This is something I've thought a lot about recently. 'cause like even, even us , in my department we've cut back on the junior people we hire. But looking across, in kind of go to market side, , I've seen people , kind of hiring less of maybe like SDRs, the people who would just be slamming phones all day long and, and sending emails. And they've been exchanging that for maybe a person who can do some of the agentic automation, but also can. Do early sales cycle and can be on the phone and can talk to people because you still need that exception path to a human. And so you need the person who can take that exception path, but who can also do the automation if you have someone who can own that entire flow, that's so powerful. And it might be, you need less people in that role, but you need, stronger people who are, be more expensive probably, or like. We're definitely knocking, getting rid of all the junior people. Roman: Dude, it, it, we're all just gonna uplevel. It's just, it's what happened with every generation of technology. I do think [00:30:00] that, we're gonna hit a point where I. There's gonna be significant upleveling that's gonna have to happen in a, in a shorter time span than has ever happened before. And that transition period is going to be interesting. But it's something that humanities have been through many times and I'm sure we will. This is not the last time we'll go through it. Jeff: Exactly. So junior folks, there you go. Go start building on your own. We have a bunch of examples earlier to go check out and start doing but just get your hands dirty and, uh, start building like I said, I could go for hours of Roman. I always have a blast whenever I get to chat with you. But. We, we do have to get back to actually our, our real jobs or, or else, junior people with AI are gonna take them away from us. So, it's been a trip, man. Thank you so much for joining and being the first on the show. This has been really fun. I'm really stoked about this. Roman: pleasure appreciate you as always Jeff. Have a good one y'all. Jeff: All right, chat soon, bud. Thank you.