Descriptions:
David Ondrej tests Gemini 3.1 Pro inside Anti-Gravity, Google’s multi-agent IDE, using it to build an autonomous web scraping agent while benchmarking the model against Claude Opus 4.6 and GPT-5.3 Codex. The benchmark breakdown covers three evaluations: ARC-AGI 2 abstract reasoning puzzles (Gemini scores 77%, ahead of both competitors), GPQA Diamond scientific knowledge questions (94%), and BrowseComp agentic search tasks (85%), with Gemini leading on all three metrics in this comparison.
The build walkthrough centers on Anti-Gravity’s agent manager, the IDE’s primary differentiator from Cursor or Windsurf — a dedicated interface for dispatching, monitoring, and orchestrating multiple AI agents working in parallel on the same project. Ondrej uses Gemini 3.1 Pro to autonomously write, test, and debug the application, integrating Firecrawl for web scraping (free tier includes 500 pages; promo code “David” adds 1,000 credits). The video captures the agent opening and controlling a browser without user input to test its own code — a concrete demonstration of agentic autonomy in practice.
The video doubles as a setup guide for Anti-Gravity newcomers, covering installation, Google account authentication for free Gemini access, agent manager configuration, and how to import settings from Cursor or VS Code. Firecrawl sponsors a portion of the content.
📺 Source: David Ondrej · Published February 24, 2026
🏷️ Format: Review
