From Hardware Engineer to AI Product Leader: Building Prototypes After Hours at Google
What was your path into product management? You started as a hardware/storage engineer.
My education background was in engineering - mechanical engineering for my bachelor's degree and electrical engineering with computer science for my master's. I think the core of product management is personality - I like to solve problems. Engineering also solves problems, but product managers solve them differently. I enjoy talking to people and doing things beyond just coding.
In 2010, I did my first startup in New York with some friends. It was similar to Yelp, bringing merchandise online. After that, I moved to California to pursue entrepreneurship in Silicon Valley. I believe if you're an entrepreneur who wants to solve problems and care about users, you naturally fit into product management.
I started my Silicon Valley career as a network and distributed hardware system engineer while getting my visa. Simultaneously, I was running an investment consulting company. I couldn't stop myself from doing new things! I've transitioned from network storage to voice AI, then to AR/VR, smart assistant, RM, and now I'm working on Gemini and agent technology. My curiosity drives me to keep learning and building new things.
You have been a product leader working in AR/VR/Android Auto space over the last 10+ years. What are some core domain skills for product managers to thrive in this hardware space?
Getting a job and doing the job are different challenges. First, you need the opportunity, which is why networking and connections are important. Once you're in, it's about learning skills.
When our AR company was developing glasses, we faced optics design challenges involving physics. For smart speakers, we encountered paint retention issues with curved designs. I didn't know these domains, but I learned by talking to people and doing research. You must be resourceful - I've used my network to find experts from companies like Alexa to help solve complex problems.
While having an engineering background might help you get a hardware PM job, once you're working, being a great learner with engineering thinking is what matters. At companies like Google, people pivot between completely different teams and domains frequently.
What does a typical day look like for you as Product Lead in Gemini-powered Gmail copilot?
It depends on the project phase. For launched products, we focus on growth hacking, discovery, education, and driving repeat usage. For early-stage projects like workflow agents, we talk to users extensively to determine direction and iterate products.
A typical day starts with checking emails and planning my top 3-4 focus areas. Then come the meetings - a crazy amount of them! I try to reserve at least 30 minutes, ideally an hour, to play with data, test our products, and analyze losses. For Gen AI products, quality is key.
About 30% of my time is data-related, 40% involves leadership communication and corporate overhead, and the rest is working with my team to solve problems and ensure projects run smoothly. At peak times, I've had several VP reviews daily, which requires significant preparation.
What's the biggest challenge you encounter right now working in the Gen AI space?
Quality is a major challenge. Businesses have much higher expectations on accuracy than consumers, so our quality bar must be higher. But LLMs have inherent randomness and creativity - they hallucinate by nature, which is actually a feature, not a bug.
When developing for businesses, many times you have to limit the LLM's power by restricting what it can do, like telling a child they can't do anything, leaving them confused sometimes.
Another challenge is the industry's rapid pace. If your MVP can't reach the market quickly enough, it becomes irrelevant. This creates particular problems for larger companies with high standards. We're constantly exploring ways to get insights earlier in the process.
Data challenges are also significant. While math and coding problems are well-defined, tasks like email writing or customer support are open-ended. How do you define success, especially in multi-step agent systems? Building evaluations is expensive, and you need specific data for each one.
Gmail's billions of users worldwide present another challenge - your feature must solve cases for everyone from soccer moms to business professionals. It's easier for startups to focus on specific personas.
Bonus Question: What's your favorite clothing brand?
I like more casual clothes. Brands like Uniqlo, Lululemon, and Everlane offer casual, comfortable, and simple clothing that I enjoy. When I first moved to California, I used to wear formal shirts, and my manager thought I was interviewing somewhere else! I have a collection of shirts and jackets that I rarely get to wear now - I don't want to be the oddly formal person at the office.
Spicy Question: You mentioned in the Product Pub panel that you love using Cursor, sometimes coding late into the night. How did you decide what to work on as a side project?
I'm a builder at heart - I prefer creating things over just talking about them. With coding agents, it's become much easier for me to build prototypes despite my deteriorated coding skills. I've built original prototypes for multiple key projects at Gmail and Google Assistant, lobbying to get engineers to integrate them into our products.
I work on my projects after my kids go to bed, typically from 10 PM to 2 AM. That's my "me time" for coding. If I face a problem at work that I can't immediately convince others to address, I prefer to build a prototype to show it's possible. This approach is more effective than a PRD full of bullet points.
I believe all PMs should build prototypes in this new AI era because PRDs may be obsolete. I spend average 2-3 hours “coding” daily and consider myself a "vibe coder." If you want to productionize the project, remember to bring in your real engineer friend before Cursor makes your code base unmanageable:)