Antigravity vs Claude Code vs Codex

追加: 2026-06-03

[00:00:00]One tool made me feel like a manager.

[00:00:02]The other made me feel like a craftsman.

[00:00:04]And a few months later, a third tool arrived that doesn't fit either box. It makes you feel like a dispatcher. For a couple of years, AI coding mostly meant chat. Describe a problem, the model spits out code, you paste it into your editor. Useful, but always slightly disconnected from how software actually gets built. Late 2025 was when that changed. I wrote the first version of this post comparing two tools that felt like a genuine fork in the road. Google anti-gravity, which treats development as a set of missions you hand to autonomous agents that plan, execute, and verify. VS Code plus Claude Code, which embeds a highly capable agent into your terminal and editor, while keeping you firmly in control. To feel the tradeoffs instead of theorizing about them, I built the same small app in both. A Pomodoro timer paired with a simple Kanban board. Then 2026 happened.

[00:00:57]The free lunch I was enjoying got smaller. The models underneath everything got better, and OpenAI's Codex matured into a real third option with a philosophy all its own. So this is the updated map. The same hands-on comparison refreshed plus where Codex fits.