Google Antigravity

What It Is

Google Antigravity is an agent-first IDE that takes a fundamentally different approach to AI-assisted development. Rather than adding AI features to an editor, it builds the editor around autonomous agents. The Manager view lets you orchestrate multiple AI agents working in parallel across different parts of your codebase, while the Editor view provides a more traditional hands-on coding experience. It's powered primarily by Gemini 3.1 Pro with support for Claude and GPT-OSS-120B.

Why It Matters

The Manager view concept is genuinely novel. No other IDE lets you spin up multiple autonomous agents, assign them different tasks, and watch them work in parallel across your workspace. It's the closest thing to an AI development team operating inside your IDE. The inclusion of GPT-OSS-120B also makes it the only major IDE offering a meaningful open-weight model option alongside proprietary ones.

Key Developments

• Mar 2026: Pricing backlash as Google shifted from free to credit-based pricing ($25 for 2,500 credits). Quota management described as unpredictable.
• Mar 2026: Model support expanded to five LLMs including Claude Sonnet 4.6, Claude Opus 4.6, and GPT-OSS-120B.
• Jan 2026: Reports of model performance degradation with weaker models and shorter context windows replacing high-reasoning variants.
• Late 2025: Security researchers documented critical vulnerabilities including remote command execution via indirect prompt injection.
• Nov 2025: Announced alongside Gemini 3 launch in public preview.

What to Watch

The concept is ahead of the execution. Watch whether Google stabilises the reliability issues (frequent crashes, quota bugs) and addresses the security concerns around prompt injection in autonomous agents. The pricing model needs to settle before teams can plan around it. If Google gets the reliability right, the Manager view could define what agent-first development looks like.