Engineer at Google DeepMind leading development of Antigravity, the company’s agent-first IDE. Previously founding team member at Codeium (acquired by Google for $2.4B) where he scaled the product from zero to over 1M users and a $1.25B valuation. Princeton CS graduate with expertise in AI-powered developer tools, autonomous agents, and building products for frontier models.
Background & Expertise
Kevin Hou joined Google DeepMind in mid-2025 after being part of Codeium’s founding team, where he served as Head of Product Engineering. Before that, he was Tech Lead Manager at Nuro (autonomous vehicles) and worked at Airbnb and Salesforce. He holds a Computer Science degree from Princeton University with certificates in entrepreneurship and statistics/machine learning.
Key Technical Areas:
- Agent-first development environments and agentic IDEs
- Code intelligence systems and embeddings for code search
- AI-powered developer tools at scale
- Full-stack engineering across web, mobile, and ML systems
Creative Work: Beyond software, Kevin is an active photographer (clients include Princeton University, Nuro AI, Psycho Bunny), woodworker, and blacksmith, bringing a maker’s mindset to building developer tools.
Notable Articles & Talks
- Introducing Windsurf Editor - Launched the world’s first AI agent-powered code editor
- AI Engineer Summit NYC 2025 - How Windsurf writes 90% of your code
- Embeddings are Limiting AI Agents - AI Engineer World’s Fair 2024
- Founder Mode: Scaling Codeium 0 to 1M Users
- Syntax Podcast: Windsurf & AI Coding - Forking VS Code to compete with Cursor
Google Antigravity IDE
Antigravity is Google DeepMind’s agent-first IDE that inverts traditional development paradigms. Built from acquisition of Windsurf team ($2.4B), delivered in 4 months, powered by Gemini 3 Pro with 76.2% on SWE-bench Verified.
Three Core Surfaces:
- Agent Manager - High-level orchestration dashboard managing multiple autonomous agents, with inbox for agent-human communication
- Agent-Controlled Editor - VS Code fork serving as tool for agent execution
- Agent-Controlled Browser - Chrome under agent control for research, context retrieval, and web interaction
Design Philosophy: Agents live outside your IDE, coordinating across surfaces. The editor and browser are tools for the agent manager, not the other way around—inverting the traditional developer-centric paradigm.
The Age of Artifacts
Antigravity introduces artifacts - dynamic representations agents generate for self-reflection, user communication, and cross-account sharing. Artifacts become interactive verification environments supporting multimodal feedback (comments, highlighting, image-space interaction). The agent decides what artifacts to create, why, and who needs them. This shifts interaction from text-based code review to visual and multimodal verification.
Building for Frontier Models
Antigravity was engineered specifically for Gemini 3’s capabilities: advanced reasoning, sophisticated tool use, reliable instruction following, long-running task support, and multimodal processing. The product embraces “deep research” phases where agents consult documentation (design docs, Slack threads, runbooks) before coding—extending agent work far beyond traditional code editing.
Kevin’s Vision: “Antigravity will be the most advanced product on the market because we are building it for ourselves.” Google DeepMind engineers dogfood it daily, and development is tightly coupled with the Gemini team—gaps in product or model inform research directions bidirectionally.
Summit Context: As the closing keynote of Day 2, Kevin’s presentation synthesized the summit’s themes on agent orchestration, multimodal reasoning, and the architectural insight that “environments are the universal abstraction.” He candidly acknowledged deployment constraints: “On behalf of the team, I want to apologize for our global chip shortage.”