From Game Dev to Google: Agentic AI, Zero to One, and the Future of Product Management

What does it actually take to build a product when no one believes it’s possible? In this episode of Product Talk, hosted by Thomson Reuters Director of Product Vinay Aradhya, Google Group Product Lead Pawel Siarkiewicz speaks on what zero to one product building really looks like, how agentic AI is rewriting the rules of software development, and what separates great product managers from the rest. Drawing from his journey as a game developer, FinTech CTO, and the lead PM who helped build Google Stadia from scratch, Pawel brings rare perspective on the craft of building and leading in an era where everything is changing fast.

Join us for new conversations with leading product executives every week. Roll through the highlights of this week’s podcast below, then head on over to our Events page to see which product leaders will be joining us next week.

Show Notes:

1. Your input bandwidth is a straw. No matter how much AI summarizes for you, your brain can only absorb so much. Give yourself time when entering a new company or domain — learning takes months, not weeks.
2. Know how you learn and advocate for it. Whether you’re a big-picture thinker or a hands-on tinkerer, understanding your own learning style and asking others to meet you there is a career-long superpower.
3. Metrics are not insight. Numbers can show you what is happening, but only talking to real users can tell you why. User research isn’t optional — it’s the difference between building the right thing and building fast in the wrong direction.
4. Don’t take users at face value — dig deeper. Users will tell you what they think they want, but the real job is understanding the underlying need. The “Mom Test” principle applies: what’s the problem behind the ask?
5. Agentic development is software where most or all of the code is written by AI agents, with humans orchestrating rather than executing. The human role shifts from writing to specifying, guiding, reviewing, and governing — which requires entirely new skills.
6. Not all code is equally suited for AI agents. Well-understood, lower-stakes tasks are great candidates. Core architecture that has never existed before and can’t fail requires far more human judgment and oversight.
7. Tokens cost money — agent sprawl is a real budget risk. Leaving agents to run overnight without guardrails can generate surprising bills. Understanding when to steer versus when to let agents run is a skill you build through experience.
8. Code drift is one of the biggest risks in long-running agentic systems. As agents hand off work to each other over time, probabilistic outputs can slowly pull a codebase in the wrong direction. Long-term guardrails that go beyond syntax and unit tests are essential.
9. Knowing when an agent is stuck versus one turn from a breakthrough is an art, not a science. Some of the most expensive mistakes come from letting agents loop endlessly. Learning when to discard a branch and start fresh is one of the most valuable skills in agentic development.
10. Massively parallel agent systems — with 50 to 100 agents working simultaneously in layered roles — are already producing results. About 25 to 30% of generated code gets discarded by higher-level review agents, but the overall output is extraordinary. Think of it like ants building an anthill.
11. Agents can evaluate and debug their own performance. You can ask an agent why it missed something, have it analyze its own context, and ask it how to improve the instructions. Software that self-debugs is no longer theoretical.
12. For human-dependent quality judgments — like whether an animation feels smooth — you still need a human in the loop. There are things a skilled designer can feel at the level of milliseconds that no agent can currently evaluate on its own.
13. The zero-to-one transition point is one of the most critical and most commonly missed moments in product development. There is a specific inflection — from proof of concept to scale-ready — where the skills, mindset, and user base all need to shift. Scaling before you’ve genuinely hit that point will cost you more than waiting.
14. Your first users are your biggest fans. Your scale users don’t care about you at all. The Stadia team learned this the hard way when they moved from enthusiastic early studios to mainstream developers who just wanted to get on with their work. Being ready for the indifferent user is the real test.
15. Time to first feedback is the most important metric in zero to one. Get something — even janky, even imperfect — into users’ hands as fast as possible. You are not building the table stakes yet; you are still building the table.
16. “AI product manager” is a buzzword that misses the point. AI is a tool like any other. Just as no one calls themselves a keyboard product manager, the core job hasn’t changed — it’s still about why you’re building, what outcome you’re after, and whether it’s the best investment of your team’s resources.
17. Every PM today should be able to use any AI system fluently. A screwdriver is a screwdriver whether the handle is blue or green. The mechanics of how to think with, prompt, and steer AI tools are now baseline expectations, not differentiators.
18. The best product managers are the unreasonable people in the room. When engineers say it’s too hard or it’ll take too long, the PM’s job is to hold the vision with conviction — not give up at the first no. Someone on the team has to be the relentless believer, and that job belongs to the PM.
19. PMs are the guiding light when everything is on fire. When the launch is a day away and the product has a million bugs, the product manager’s job is to be the calm, steady, optimistic center. Engineers can afford to be cynical. PMs cannot.
20. Embrace the blur between engineering and product — but don’t forget your superpower. Learn to build prototypes, get comfortable with code, and explore vibe-coding platforms. But if your edge is user empathy, storytelling, and team connection, protect that. Exceptional people who are equally strong at both are rare for a reason.