In this video, I went hands-on with Google’s brand new Antigravity IDE and put its AI agents to a real test by vibe-coding a complete AI News Brief app. I walk you through the first-time setup (themes + agent modes), why I prefer the agent-assisted mode, and what the UI feels like—left side very VS Code, right side very Cursor. I also break down the built-in model options (Claude Sonnet / Gemini Pro) and explain what “low vs high” really means: speed vs reasoning. Then I do what matters: I define requirements in a markdown file, wire an API key via an .env file, and use Antigravity’s plan mode to review the implementation strategy before code generation. The plan is solid—components, folder structure, service integration, and even a testing strategy leveraging Antigravity’s bundled Chromium browser. But in execution, the agent struggled with terminal commands that need selection, got confused around .gitignore, and couldn’t correctly align the env var naming—so I had to fix the API key manually. After that, the agent retested in its sandboxed browser, validated cards/tiles rendering, and even generated a test recording. Verdict: Antigravity is free (for now) and exciting, but for real-world coding reliability and “just works” workflow, I’m still sticking with Cursor 2.0 today—and I’ll re-evaluate as these IDEs evolve.