Google's Antigravity (Fully Tested): Does Google's New AI Code Editor hold up?

Key Concepts

• Anti-gravity: Google's new AI-powered coding editor.
• Wind surf: An existing AI coding editor, whose technology and team were acquired by Google.
• Gemini 3: The AI model powering Anti-gravity.
• Agent: An AI assistant within the editor that can perform coding tasks.
• Agent Manager: A new interface in Anti-gravity for managing AI agents and their tasks.
• Proprietary Context Engine: A feature of Augment Code that provides relevant code snippets quickly.
• Remote Agents: A feature of Augment Code for managing pull requests on cloud workers.
• Agentic Harness: The underlying system that enables AI agents to perform tasks.
• Vibe Coding: A term used to describe AI coding that feels superficial or lacks deep understanding.

Google's Anti-gravity: A Deep Dive into the New AI Coding Editor

Google has launched "Anti-gravity," an AI-powered coding editor that is essentially Wind surf with Gemini 3 integrated. This new experience aims to redefine AI editing by offering features like autocomplete, an agent, and a dedicated agent view for managing these agents. The interface draws inspiration from existing tools like Cursor and Verdant, but users have noted some peculiar aspects upon initial use. Anti-gravity is available for Mac, Windows, and Linux, and during installation, it offers to import settings from Wind surf, Cursor, and other similar tools.

Sponsorship: Augment Code

Before delving deeper into Anti-gravity, a word from the sponsor, Augment Code. Augment Code is presented as an enterprise-grade AI solution designed for large, fast-paced codebases, distinguishing itself from "toy apps" or "vibe coding." Its key differentiator is a proprietary context engine that delivers millisecond-relevant code snippets, even across massive monorepos (up to 100k files). It feeds the entire repository, potentially millions of lines of code, into the best available AI model in real-time. Augment Code offers smart, in-context suggestions tailored for production code, leveraging models like Claude 3.5 Sonnet. It automatically upgrades to the best model without user intervention and integrates seamlessly with popular editors such as VS Code, JetBrains, Vim, and Cursor, without requiring forks or compromises. Security is a priority, with code never being used for training and support for customer-managed encryption keys. Pricing is based on successful requests (pay-per-message), avoiding seat licenses or complex token calculations. Recent features include remote agents, enabling the launch, monitoring, and merging of pull requests from parallel cloud workers without taxing local CPU resources. A 14-day free trial is available at augmentcode.com.

Visual and Functional Similarities to Wind surf

Upon examination, Anti-gravity bears a striking resemblance to Wind surf. The visual cues, particularly the colorful file icons, are a direct inheritance from Wind surf. This similarity is attributed to Google's acquisition of Wind surf's technology, code, and key personnel, including its founders and CEO, who have since joined Google DeepMind. This acquisition means the core technology and the team behind it are now operating within Google.

The user interface of Anti-gravity closely mirrors older versions of Wind surf, predating Cognition's acquisition of Wind surf and subsequent changes to its product. This is evident in the settings panel, which is virtually identical in both editors. Even the tooltip messages for features like "autofix lint errors" are the same, with only minor name changes (e.g., "cascade" replaced with "agent"). The customization pages also exhibit the same layout and functionality as Wind surf.

User Experience and New Features: The Agent Manager

The primary method of interaction with Anti-gravity is through its editor interface, which opens an agent panel on the right side. Users can input prompts, and the agent will execute the requested tasks, a common paradigm in agentic editors. However, the most significant new addition is the Agent Manager.

The Agent Manager is described as a "low-budget version of Verdant," another agent management interface. While Verdant is praised for its thoughtful design and intuitiveness, Anti-gravity's Agent Manager, though similar in concept, is considered less refined. It allows users to view projects and their associated running tasks, navigate between different projects, and initiate new threads. An "inbox" feature displays pending tasks and those requiring user attention.

Performance and Benchmarking

The reviewer tested Anti-gravity with several benchmark tasks:

• Benchmark Question: The agent successfully created an implementation plan and executed it, although it produced two errors that required fixing. The overall experience was deemed "fine."
• Go TUI Calculator: The agent performed well but also generated one error that needed manual correction.
• Godo Game: This task proved problematic. Despite five prompts (exceeding the reviewer's usual three-prompt limit), the agent failed to complete the task.
• Spelt Benchmark: This long-running task also resulted in failure, with the agent erroring out after a period.
• Nux App and Tari App: Similar to the Spelt benchmark, these tasks also failed due to the agent's inability to handle long-running operations.

The reviewer concludes that Anti-gravity is currently "extremely buggy," and the agent harness, even with Gemini 3, does not perform well, often making things worse.

Browser Integration and Agent Harness Limitations

A feature highlighted as potentially new is the agent's ability to use a browser to verify task completion. However, the reviewer dismisses this as "nonsense," noting that similar capabilities have existed in tools like Klein (with Claude 3.5 Sonnet's browser support), Stagehand, and MCP Dev Tools for some time. The reviewer's experience with this feature has been largely negative, as the agent often reports success based on UI appearance, overlooking significant UI issues or outright errors.

The core issue identified is the agent harness, which is deemed inadequate. It appears to prioritize token saving, leading to suboptimal results. The overall user interface is described as "super bolted on" and lacking cohesion, feeling akin to using Pair AI or "vibe coded with 3.5 Sonnet."

Critical Assessment and Redundancy Concerns

Despite having "extreme respect for Google DeepMind," the reviewer states that Anti-gravity is "not a quality product" and should have been delayed. A significant concern is the perceived redundancy of the product. The reviewer questions its necessity, pointing to existing Google offerings:

• Firebase Studio: Offers a similar bolt-like interface and VS Code integration.
• Gemini Code Assist VS Code Extension: Provides autocomplete and agentic capabilities.
• Gemini CLI: Offers similar functionalities.

While acknowledging that these tools might offer free credits for using Gemini 3, the reviewer would not pay for Anti-gravity in its current state, nor would they use it for free.

Conclusion

In summary, Google's Anti-gravity is a rebranding and integration of Wind surf technology with Gemini 3. While it introduces an Agent Manager, its core functionality and user interface are heavily derived from Wind surf. The product suffers from significant bugs and performance issues, particularly with complex or long-running tasks. The agent harness is considered weak, and features like browser integration are not seen as innovative or effective. The reviewer questions the product's necessity given existing alternatives, both within and outside of Google. Despite its potential for accessing Gemini 3, Anti-gravity is not recommended in its current state due to its unpolished and buggy nature.