# Pavan Kumar: full site text Source: https://pavank.ai --- # Pavan Kumar AI product leader and entrepreneur, building multimodal GenAI products. PM Lead at Google Labs, building AI experiences with world models and agents. 15 years of experience building consumer and enterprise AI products used by 100M+ MAUs. ## Key affiliations Google, Apple, UC San Diego, Stanford, Cocoon Cam, NetApp. ## Published in TechCrunch, TechRadar, Forbes, CNBC, Inktalks, Samsung NEXT. ## Explore - [About](/about) (plain text: [/about.md](/about.md)) - [Work](/work) (plain text: [/work.md](/work.md)) - [Writing](/writing) (plain text: [/writing.md](/writing.md)) - [Studio: chat with my AI twin](/studio) (plain text: [/studio.md](/studio.md)) - [Contact](/contact) (plain text: [/contact.md](/contact.md)) --- # Studio: Ask My AI Twin Chat with Pavan Kumar's AI twin at [/studio](/studio). Get insights on building AI products, career advice, and more. The AI twin is grounded on Pavan's writing, work, and public interviews. It answers as a smart friend: direct, warm, and slightly playful. ## Prefer to connect directly - Email: pavankumar.pn@gmail.com - LinkedIn: https://linkedin.com/in/pkpn - X (Twitter): https://x.com/pk_notes --- # Products & Impact 0-to-1 products across AI, video agents, world models, and applied ML. Shipped to millions of users and $MMM revenue. ## Project Genie with Street View (2026-present) Grounded a generative world model on real-world imagery from Street View, so you can explore and remix any location in the US as a 3D scene. Went viral on X at launch. - Learn more: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/project-genie-expands/ - Demo: https://www.youtube.com/watch?v=bxv4IkobUPI ## Google Beam Video Agents (2026-present) Founded video agents on Beam to explore the future of human-AI interaction. Walk up and talk to a life-size, multimodal agent using tools like Search and Maps in real time. Debuted at Google I/O 2026. CNET called them "the realest almost-people." - Article: https://www.cnet.com/tech/services-and-software/google-io-beam-telepresence-ai-agent-demo/ - Video: https://www.cnet.com/videos/is-googles-uncanny-virtual-human-a-future-coworker-or-concierge/ ## Portraits: AI Twins of Experts (2024-present) Co-founded a 0-to-1 product: talk with AI twins of real experts, grounded on their entire body of work. Led product across voice, LLMs, and RAG. Launched with Kim Scott and YouTube creators. - Learn more: https://blog.google/technology/google-labs/portraits/ ## Dynamic Layouts for Google Meet (2022-2024) AI-native Meet layout using face tracking, speaker detection, and diarization. Gives everyone a fair seat at the table. Shipped to millions globally. - Learn more: https://support.google.com/meet/answer/15929512 ## Studio Look: AI Video Enhancement (2022-2023) Meet's first generative AI feature, enhancing video quality in real time. Shipped with cloud and on-device inference. Reached millions globally. - Learn more: https://support.google.com/meet/answer/13948742 ## Google Beam: 3D Telepresence (2022-2024) Led ML and cloud infra for real-time volumetric capture and neural rendering. Drove Meet integration and product vision. - Learn more: https://blog.google/technology/research/project-starline/ ## Google Cloud HPC & ML Infrastructure (2021-2022) First PM for Cloud HPC. Launched HPC Toolkit, Titanium RDMA, and HPC VM images. Now used to train LLMs. Led 5 products and 50+ people. $MMM revenue impact. - Learn more: https://cloud.google.com/cluster-toolkit/docs/overview ## Cocoon Cam: Computer Vision Baby Monitor (2014-2020) Co-founded and led product and engineering from day zero to Series B and acquisition by Alarm.com. Built the world's first computer vision contactless vital monitor: hardware, AI vision models, and mobile apps. - Learn more: https://jacobsschool.ucsd.edu/news/release/2251 --- # About Pavan Kumar Hi, I'm Pavan Kumar, PM Lead at Google Labs. I am currently focusing on AI experiences with world models and AI agents. I work closely with teams across Google, including DeepMind, YouTube, and Workspace, to bring new AI into products that millions of people use every day. I've also shaped ML strategy for Google Beam and Google Meet, focusing on applied AI, data, partnerships, performance, and fairness. Before this, I built supercomputing products at Google Cloud that power some of the most demanding ML workloads on the planet. What I love most is taking an idea from zero to one: from a first sketch to launch and real growth. That drive led me to co-found Cocoon Cam and lead both product and engineering. We built the world's first computer vision and AI-powered contactless vital monitor from the ground up, including the hardware, AI vision models, and Android/iOS apps. I also co-founded Chrysalis Cloud to help developers build video-based products. I believe multimodal generative AI can change how we learn, create, and solve hard problems. Over the years, I've shared this work at events like the first-ever White House Demo Day hosted by President Obama, and delivered the guest commencement address at UC San Diego's master's graduation ceremony. ## Speaking Highlights - Samsung Next Keynote (2019) - UCSD Commencement Speech (2018) - White House Demo Day (2016) ## What I Bring to the Table I love products that sit at the intersection of deep technology and real human need. My background is a mix of technical depth, entrepreneurial scrappiness, and product strategy. - **Shipping AI products:** Track record of building consumer and enterprise AI products that people actually use. Consumer: Cocoon Cam, Project Genie, Portraits. Enterprise: Cloud AI Vision, Cluster Toolkit for large-scale distributed training, Google Meet AI, Beam Video Agents, and Google Beam. - **Zero to one:** I thrive taking an idea from first spark to real impact, whether co-founding a startup from day zero or being the first PM on a new Google product. Comfortable navigating ambiguity, building conviction, and shipping launches that matter. - **Product strategy:** I think from first principles and adapt to the product lifecycle. Early on, I run lean experiments to find product-market fit and get from 0 to 1. At scale, I focus on distribution, operational discipline, and the systems needed to reach 100M+ users. If you are building at the frontier of AI and want to make something real, let's connect. ## Personal Life I live in Mountain View, California, with my wife, Anusha. We are outdoor people: Bay Area trails, scuba, and travel whenever we can. We are vegan/vegetarian and always on the hunt for our next favorite dish. --- # Get in Touch Ready to turn AI ideas into real products? Let's talk. - Email: pavankumar.pn@gmail.com - LinkedIn: https://linkedin.com/in/pkpn - X (Twitter): https://x.com/pk_notes ## Great conversation starters - "I'm building an AI tool that helps with X, but struggling with Y..." - "We have an MVP but need help scaling the AI infrastructure..." - "I have this idea for multimodal AI but not sure if it's feasible..." - "Looking for a technical co-founder who gets AI product strategy..." --- # My Favorite AI Products I Actually Use as a PM: 2026 Edition Published 2026-06-30. 5 min read. Tags: AI Tools, Productivity, 2026. > An updated look at my PM toolkit as agentic AI starts retaining context, running on schedules, and handling multi-step work on its own. Last year, I shared [the AI products](https://pavank.ai/writing/my-favorite-ai-products-i-actually-use-as-a-pm) that actually fit into my daily PM workflow. For 2026, the bar stays the same; I'm sharing AI products that help me make decisions, build artifacts, and prioritize. The notable shift this year is the practical integration of agentic AI. These tools now retain context, operate on schedules, and handle multi-step tasks autonomously. This matters for PMs because our job is to bring clarity to messy problems, align the team, and keep decisions moving. ## 1. Antigravity: My go-to workspace for work Antigravity is my go-to for work. I've loaded all my project context into it, so I don't have to keep re-sharing context every time I start a new task. It also connects to my workspace tools and has the harness to actually do things. I use it for writing, planning, and managing both professional and personal projects. For any major initiative, I set up a workspace and feed it the relevant context: documents, meeting notes, emails, chat logs, and rough ideas. I rely on it to draft PRDs, build project dashboards, write leadership updates, and organize scattered notes into a cohesive plan. Claude handles deep reasoning and coding, while Gemini 3.5 Flash is great for quick summaries and fast drafting. **Try this:** Set up a workspace for one active project. Load in your documents, emails, and meeting notes once, then let the agent work across them without repeating context. Ask it to take action through integrations with your calendar, email, and chat. ## 2. Gemini Spark: Automated research and synthesis I started the year setting up a local machine with OpenClaw to run scheduled agent tasks. Gemini Spark feels like a hosted version of that, with more predictability, a better interface, and tighter control. It connects to my data sources like Workspace, so it can pull from email, calendar, and docs to run things like daily digests. The daily digest flags pending decisions, missed messages, and threads where I might be the bottleneck. The weekly digest surfaces follow-ups and unaddressed notes before the weekend. **Try this:** Schedule a Friday review asking the AI to flag tasks you avoided, pending choices that would clarify next steps, and conversations needed to keep the team moving. Use this to decide what to finish, delegate, or drop. ## 3. Lovable: Rapid prototyping Lovable is my go-to for quick prototypes, side projects, and my personal website. I can turn a rough idea into a clickable interface before pulling an engineering team into it. A tangible prototype changes how you evaluate an idea. Once it is interactive, it is much easier to spot friction points, identify gaps, and get meaningful feedback. Lovable also provides excellent default UX and visual design, alongside strong new security features. **Try this:** Before writing a comprehensive PRD, generate the most basic clickable version of your idea. Use it to refine the user problem, test the core interaction, validate your assumptions, and identify what to cut. ## 4. ChatGPT: Everyday planning and research I use ChatGPT for trip planning, general research, cooking, and organizing my personal life. I recently purchased a MUS-01 watch by relying entirely on a single ChatGPT thread to handle brand comparison, research, and my final choice. I also rely on it for travel itineraries, vegan recipe ideas, and side projects. Features like memory, folders, image generation, and deep research make it increasingly useful over time, as I spend less effort explaining my constraints. It even successfully compiled my scattered recipes into a formatted cookbook. **Try this:** Use a single chat thread as a decision journal for a major purchase or trip. Log your constraints, tradeoffs, and final choice, then revisit the thread later to see if the outcome matched your original priorities. ## 5. Portraits: Interactive coaching on demand I use Portraits when I need targeted coaching on a specific situation. The two personas I return to most often are Matt Dicks for storytelling and Kim Scott for navigating leadership conversations. Portraits is incredibly helpful for rehearsing difficult conversations, refining a narrative, or working through a real-world problem using a trusted framework. It transforms static advice into a dynamic coaching session. **Try this:** Before a difficult conversation, write down the message you naturally want to send. Have the AI coach you by generating three alternatives: one that is exceptionally clear, one that is highly empathetic, and one that addresses the hard truth directly. ## 6. Project Genie: The shift toward interactive media I recently became the PM for Project Genie, so I am close to it, but I genuinely think it represents a meaningful evolution in AI. Genie focuses on world models and simulators that can change how we interact with media. Most consumer media today is passive: we watch, read, and scroll. World models turn these experiences into environments you can explore and influence, similar to the Street View grounding work we recently launched. **Try this:** Take a piece of passive media you enjoy, like a travel vlog, learning module, game trailer, or map. Consider how the experience would improve if you could actively move through it and alter the environment. ## At a glance | Product | What I use it for | Why it helps | | :--- | :--- | :--- | | **Antigravity** | Writing, planning, project work, dashboards, updates | My go-to workspace; holds all project context and connects to my tools. | | **Gemini Spark** | Scheduled research, daily synthesis, weekly review | Flags pending decisions and follow-ups at the right time. | | **Lovable** | Prototypes, side projects, personal website | Turns concepts into clickable interfaces quickly. | | **ChatGPT** | Trip planning, research, shopping, cooking, hobbies | Adapts to personal preferences to help with daily choices. | | **Portraits** | Storytelling, leadership prep, targeted coaching | Provides interactive coaching using specific leadership or communication frameworks. | | **Project Genie** | Interactive experiences via world models | Demonstrates how passive content can become explorable. | Canonical page: https://pavank.ai/writing/my-favorite-ai-products-i-actually-use-as-a-pm-2026-edition --- # My Favorite AI Products I Actually Use as a PM: 2025 Edition Published 2025-09-07. 5 min read. Tags: AI Tools, Productivity, 2025. > Five AI products that actually made it into my daily PM workflow in 2025. > Also see the [2026 edition](https://pavank.ai/writing/my-favorite-ai-products-i-actually-use-as-a-pm-2026-edition) of this list. Instead of using one AI tool to do everything, I've been curating a small set of tools that are exceptional at specific jobs. Here are the five products that have become essential to my work, what I use them for, and why they are special. ## 1. Favorite for Deep Work: Gemini + Google AI Studio For the heavy lifting of product management work like creating strategy docs, PRDs, Gemini is my go-to. It excels at handling massive, messy context without losing the plot. My workflow involves feeding a mix of UXR transcripts, previous PRDs, and backlog notes into a single AI Studio chat. It synthesizes this information into a coherent strategy document that includes risks and open questions. **Try This:** Drop your user research notes and two related strategy docs into AI Studio. Ask it, “Draft a two-page brief for our next project. Surface the three biggest risks and propose a small experiment to de-risk the top one.” ## 2. Favorite for Life & Creativity: ChatGPT For my personal life, ChatGPT is my choice for its powerful personalization and memory. It acts as a personal assistant that learns my preferences over time. I use it for everything from escaping the blank page on a home project to getting quick recommendations. When I plan a trip, it remembers from our previous preferences, making its suggestions relevant. On a walk, I use Voice Mode to brainstorm lifestyle hacks. It's the perfect glue for life's logistics. **Try This:** Take a photo of a room in your house. Ask for three layout variations in a style you like. Pick your favorite and refine it with more constraints, such as "keep the 72-inch sofa" or "add more storage." ## 3. Favorite for Building to Think: Lovable To move past documents and truly "build to think," I use Lovable. It is the fastest way I know to turn a concept into a functional prototype that people can click. I can describe a user flow in plain English and Lovable scaffolds the front end, connects to a Supabase backend for data, and hooks into the Gemini API for intelligence. The result is a shareable demo link I can produce in a single afternoon. **Try This:** Prototype a new feature using Lovable. Put it in front of your team for feedback before writing a single line of a PRD. ## 4. Favorite for Organizing Information: NotebookLM I use NotebookLM to turn dense materials like PDFs and user research into assets the team will actually consume. It keeps a living notebook for a project so key insights don't get lost in chat threads. I can also generate a podcast-style Audio Overview for my commute or a 90-second Video Overview to share with executives. **Try This:** After a week of user interviews, upload the transcripts to NotebookLM. Ask it to surface the top three themes and generate a short audio summary. Post it in your team’s channel for faster alignment. ## 5. Favorite for the Human Side of Work: Portraits For the essential human side of leadership, I use Portraits (I’m the PM lead on this experiment!) It is a new experiment from Google Labs that lets you have a conversation with an AI version of a trusted expert, built in partnership with them and based on their own work. The first Portrait features Kim Scott, author of *Radical Candor*. Before a critical one-on-one, I can ask the AI Kim for coaching. For example, "How can I tell a team member that their quest for perfection is slowing us down, without destroying their motivation?" It provides advice and phrasing grounded in her principles, in her voice. This preparation helps me walk into the real conversation calmer and clearer. **Try This:** Think of a tough conversation you need to have. For instance, ask the Kim Scott Portrait: "How can I tell a team member that their quest for perfection is slowing us down, without destroying their motivation?" ## My AI Toolkit at a Glance | Goal | Tool | Key Strength | | :--- | :--- | :--- | | **Write long docs & build with APIs** | Gemini + AI Studio | Best-in-class for long context, file uploads, and fast iteration. | | **Get personalized creative & life planning** | ChatGPT | Great for personalized planning, creative brainstorming, and voice capture. | | **Build a functional demo today** | Lovable | The fastest path from a simple prompt to a live URL. | | **Make research digestible** | NotebookLM | Turns dense information into engaging audio/video summaries. | | **Get coaching for tough conversations** | Portraits | AI coaching from trusted experts, grounded in their own work. | Canonical page: https://pavank.ai/writing/my-favorite-ai-products-i-actually-use-as-a-pm --- # The Team You Feel Across the Room Published 2025-09-07. 6 min read. > Spot a great team in minutes: laughter during debugging, problems seen as puzzles, and a palpable sense of "we've got this." Discover what makes them special. You can spot a great team in five minutes. There’s a lightness. You'll hear laughter during a debugging session. A teammate says, "I've got you," and you know they mean it. Demos are presented like compelling short films: tight, vivid, and with a touch of calculated risk. You'll see problems surface early, not as dead ends, but as puzzles to be solved. The conversation quickly shifts from "this is stuck" to "here are a few ways I could approach this." It’s a team of problem-solvers, not just problem-spotters. They view the scoreboard not as a threat, but as a celebration of their shared momentum. I've had the privilege of being part of teams like this. They are the ones you don't want to leave because the collective energy is invigorating. This is my attempt to capture that essence and articulate what makes these teams special. #### What a High-Performing Team Feels Like * **Calm Clarity:** A shared understanding of the single most important objective for the week. * **Pride in Shipping:** We release small, frequent updates and are proud to attach our names to the work. * **Candid & Kind:** We challenge ideas directly because we care personally about each other's success. * **Compounding Trust:** Small commitments are consistently met, making it feel safe to take on ambitious projects. * **Generosity:** We offer a hand when it's needed, knowing everyone is driving their own work forward first. * **Genuine Joy:** The team shares inside jokes and small rituals, creating a light and positive atmosphere. #### Five Principles of a "Felt" Team These five principles draw heavily from Sammy Lee's Autopilot Leadership model, which I had the pleasure of practicing in a workshop with Sammy himself and have seen firsthand that they work. The core idea is simple: build quiet guardrails and shared habits so the team moves with clarity and autonomy. 1. **Build the Right Team:** This is not about exclusively hiring "smart people." It's about assembling a team with complementary strengths, a shared bias for action, and a deep-seated belief that when the team wins, every individual on it succeeds. We value the engineer who simplifies, the designer who makes the story obvious, and the product manager who remains focused on the problem while staying flexible on the solution. 2. **Align on a Common Goal:** A successful week should have a clear headline. Every team member should be able to complete the sentence: "If we accomplish X by Friday, we will have made a meaningful improvement." This clear objective helps us eliminate work that doesn't contribute to the primary goal. True alignment brings a sense of relief and focus. 3. **Create a High-Trust Environment:** Trust is forged in the consistent keeping of small promises. It's also built in how we handle the truth. Because we've established that we have each other's backs, a tough critique on a design or a piece of code is heard for what it is: a shared desire to make the work better, not a personal attack. This allows us to be honest without being unkind. 4. **Empower Effectively:** True empowerment is providing genuine ownership and clear boundaries. We define the problem, the user, and the constraints, then trust the team to find the path. We support their decisions publicly and provide constructive feedback privately. There is a palpable sense of dignity in this approach. 5. **Coach and Develop Talent:** We are intentional about growing our people. The feedback we give is candid because we care about their growth, and it sits on a foundation of specific appreciation. We provide support when team members take on challenging responsibilities, enabling them to become effective. #### How This Applies to Building AI The world of AI is fast-paced and probabilistic. Successful AI teams navigate this ambiguity with a set of ingrained habits: * **We Debate with Data:** The phrase "Let's A/B test that prompt" is a more productive approach than "I have a feeling..." * **Quality is a Shared Responsibility:** Data labeling, red-teaming, and identifying edge cases are collaborative efforts. * **Shipping Is Prioritized Over Perfection:** We focus on getting value to users quickly and then iterating. * **The Story Matters:** We demo our work with clear examples that demonstrate the "before and after" in a compelling 30-second narrative. Strong guardrails and shared norms provide the structure necessary for creative and bold exploration. #### Final Thoughts The most exceptional teams create an environment where excellence feels like the natural, and even enjoyable, way of working. That is the kind of team I want to build, be a part of, and foster. *** #### **References** * Coyle, Daniel. *The Culture Code: The Secrets of Highly Successful Groups*. * Lee, Sammy. *The Autopilot Leadership Model*. * Lencioni, Patrick. *The Five Dysfunctions of a Team: A Leadership Fable*. * Scott, Kim. *Radical Candor: Be a Kick-Ass Boss Without Losing Your Humanity*. Canonical page: https://pavank.ai/writing/the-team-you-feel-across-the-room