Automating PRDs and Beyond: AI for Product Teams | Roman Gun, VP (Zeta Global) | LaunchPod AI
What if you never had to write a PRD again?
On this week’s episode of LaunchPod, we're excited to kick off a new show focusing on the real-world AI applications that product teams are using to become faster, more efficient, and smarter.
I’ve been traveling across the country and meeting product leaders who all have the same question: how are other product leaders using AI on their teams and their workflows?
So, in this first episode of LaunchPod AI, we sat down with Roman Gun, VP of Product at Zeta Global, to learn how he’s using AI agents to automate PRDs and turning them from static documents into living, collaborative tools that evolve with your team.
“If you find a product person that likes writing PRDs, run.”
Roman’s take on product documentation is refreshingly honest. Like many of us, he found traditional PRDs to be bloated, rigid, and outdated. So he built AI agents to fix them.
Instead of writing docs from scratch, Roman now uses a custom-trained AI agent that:
Understands his preferred PRD format
Prompts him for missing info
Knows how to structure outputs for platforms like Confluence or Jira
Continuously updates based on live conversations with his teams
Roman starts with a voice interaction (a high-level goal, or TL;DR), and lets the agent ask for clarity or missing inputs. Then it writes the doc. Later, it updates itself using team meeting transcripts.
“Now I’m having a conversation the same way I would be having a conversation with an engineer or a QA manager or a designer.”
Once the PRD is generated, it becomes a living document, revised through a “human → AI → human → AI” loop. It’s not a one-off output. It’s a collaborative workflow.
“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.”
The best part? What used to take hours now takes minutes — and it works across one-pagers, epics, weekly updates, and more.
Prototyping, Faster: From Component Library to Functional Mockup
Roman also outlined how his team uses AI to build functional prototypes from scratch. The workflow is straightforward:
Feed the AI a component library or dashboard screenshot
Describe the interface or feature you want to build
Let it generate an image or layout
Push it to Figma
Optionally convert to baseline code for developer handoff
“It gives you a real narrative tool... something you can test with users. The users don’t care that it doesn’t perfectly align with your component library.”
More than anything, this process makes abstract ideas real and fast. It helps engineers and designers get aligned early, fosters creativity, and accelerates decision-making.
“You get to that starting point of the creative act quicker.”
Orchestrators Over Executors
Zooming out, Roman and I talk about where AI is taking us, not just in workflows, but in how we think about talent and team structure.
“People’s core competency is gonna be orchestration and managing a combination of humans and agents.”
As AI takes on more rote and procedural work, teams need people who can direct agents, spot problems, connect workflows, and keep learning. It’s less about vertical depth in a single area and more about lateral range across multiple functions.
“I think we’re gonna move toward a T-shaped society.”
That shift is already showing up in hiring practices. Some companies are pulling back on junior hires, leaning on AI tools for the work entry-level team members used to do. Roman sees a risk there, but also an opportunity.
“We’re gonna have to hire a lot of new types of juniors... hybrid juniors who own an entire workflow or responsibility set.”
These will be people with curiosity, communication skills, and enough technical fluency to manage agents and outcomes. They’ll be orchestrators from day one.
Magic: The Gathering, Agents, and the Power of Play
When Roman wanted to level up his Magic: The Gathering game, he used the opportunity to experiment with AI agents and learn what they were capable of.
“I actually built an agent that would help me build my own decks... I have this thing where I don’t want to do what everyone else is doing.”
The project involved simulating four-person games, analyzing deck performance, and adapting for banned lists and real-time changes. Along the way, Roman learned about token limits, context windows, and model differences, and he brought all of that back to his day job.
“There’s no more barrier to entry to learn anything if you really want to learn it. There never was. But now it’s so, so much easier.”
The point wasn’t just to win more matches (though he did). It was to experiment with real agent-based workflows in a creative, high-complexity environment. And it worked.
The TL;DR
AI isn’t replacing your job. It’s changing what your job is. Roman Gun’s approach shows us what that future can look like — one where product leaders move faster, teams align sooner, and docs, decks, and prototypes evolve in real time.
The key is to start now, start small, and keep learning.
“Everything you do builds on itself and it’s going to get easier... Each time you’re going to get better and better and better.”
Links
Resources
Tools
Chapters
00:25 AI in Action
01:27 Real-World AI Applications
02:41 Community and AI Learning
06:12 Automating PRDs with AI
11:51 Training AI for Better Outputs
14:48 Prototyping with AI Tools
16:03 Generating Baseline Code with AI Tools
16:17 The Power of Prototyping and User Feedback
16:36 Accelerating the Creative Process
18:18 Magic: The Gathering and AI Deck Building
19:28 Challenges and Innovations in AI Deck Building
21:35 Learning and Applying AI in Real Life
25:30 The Future of Junior Roles in Tech
30:21 Final Thoughts and Encouragement
What does LogRocket do?
LogRocket's Galileo AI watches user sessions for you and surfaces the technical and usability issues holding back your web and mobile apps. Understand where your users are struggling by trying it for free at LogRocket.com.