← back
Defying Gravity - Kevin Hou, Google DeepMind
Takeaway
Antigravity reframes the IDE around an agent manager that orchestrates editor and browser agents, leveraging Gemini 3's longer-horizon multimodal tool use.
Summary
- Launches Google Antigravity, an agent-first IDE from DeepMind with three surfaces: editor, agent-controlled Chrome browser, and an agent manager.
- Agent Manager acts as a higher-level inbox/control panel for many parallel agents, with OS-level notifications and <100ms switching to the editor.
- Agent-controlled browser executes JavaScript, scrolls/clicks, and produces recorded verification videos instead of just diffs.
- Built around Gemini 3 Pro's gains in intelligence/reasoning, longer-running tasks, tool nuance, and multimodal capabilities.
- Frames product evolution as autocomplete → chat → agents, each unlocked by a model-capability step function.
antigravitygemini-3agent-ide
Original description
Why we built Google Antigravity, and discussing the future of agentic IDEs with Gemini 3. Speaker: https://x.com/kevinhou22 AIE is coming to London and SF! see dates and sign up to be notified of sponsorships, CFPs, and tickets: https://ai.engineer **Timestamps:** 00:00 Intro & Anti-gravity Launch 01:26 Anti-gravity Product Overview (Three Surfaces) 02:16 Agent Manager & Editor Integration 03:13 Agent-Controlled Browser & Context Retrieval 05:51 Philosophy: Model Capabilities Driving Product Paradigms 07:03 Four Categories of Improvements (Intelligence, Tools, Long-running, Multimodal) 08:31 Computer Use & Visual Verification 11:18 Image Generation & Design Iteration 12:26 The "Artifacts" Interaction Pattern 16:51 Artifacts for Memory, Notifications, & Feedback 20:34 The Research-Product Flywheel 23:35 Closing Remarks