Google Antigravity: The Agent-First IDE Redefining Software Development

Google Antigravity: The Agent-First IDE Redefining Software Development

December 27, 2025

Development is taking off. On November 18, 2025, Google unveiled Antigravity, an agentic development platform that marks a fundamental shift in how we develop software.

Unlike traditional IDEs or even first-generation AI coding assistants, Antigravity treats AI agents as first-class citizens – they don't just suggest code, but plan entire features, execute them across multiple surfaces, verify their own work, and learn from your feedback.

---

Table of Contents

1. What is Google Antigravity?
2. Core Architecture
3. Key Features
4. Interaction Modes & Autonomy Levels
5. Customization: Rules & Workflows
6. Agent Security Configuration
7. Changelog: December 2025
8. Rate Limits & Pricing
9. Known Issues & Bugs
10. Security Vulnerabilities
11. User Feedback & Reception
12. Performance Benchmarks
13. Getting Started
14. Tips for Maximum Productivity
15. Competitive Comparison
16. What's Next

---

What is Google Antigravity?

Google Antigravity is an AI-powered integrated development environment designed around an "Agent-First" paradigm.

Built as a fork of Visual Studio Code (through Google's $2.4 billion acquisition of Windsurf and its team under CEO Varun Mohan in July 2025), Antigravity goes beyond code completion, offering autonomous agents that can:

• Plan complex multi-stage development tasks
• Execute code changes via editor, terminal, and browser
• Verify their own work through automated tests and browser validation
• Learn from past interactions to improve future tasks

"This came from my team" — Varun Mohan (former Windsurf CEO) on Twitter

The platform is available in a free public preview for macOS, Windows, and Linux, with generous rate limits for Gemini 3 Pro for all users.

"We want Antigravity to be the home base for software development in the age of agents." — Google

---

Core Architecture: Two Views, One Mission

Editor View

The familiar VS Code-like interface enhanced with AI capabilities:

• Tab Completions for synchronous, hands-on coding
• Inline Commands for quick AI-powered edits
• Agent Sidebar for conversational interactions
• Direct file editing with AI-powered suggestions

Manager View (Agent Manager)

The revolutionary "Mission Control" for orchestrating autonomous agents:

• Spawn and Monitor multiple agents working in parallel
• Asynchronous Task Execution across different workspaces
• Real-time Progress Tracking with detailed thought processes
• Artifact Review for transparent work verification

---

Key Features That Make Antigravity Stand Out

1. Verifiable Artifacts

Delegating work to an agent requires trust. Antigravity solves this by having agents generate artifacts – tangible results that allow you to verify the agent's logic at a glance:

You can leave feedback directly on artifacts – similar to comments in a document – and the agent will incorporate your input without interrupting its execution flow.

2. Native Browser Integration

One of Antigravity's clear differentiators is its deep integration with Chrome through a dedicated browser extension.

The Browser Sub-Agent can:

• Launch your app on localhost and navigate the UI
• Click through interfaces and fill out forms
• Read console logs and debug errors autonomously
• Take screenshots and record videos of test sessions
• Validate the actual user experience end-to-end

The browser runs in a separate Chrome profile for security – no account data, cookies, or history from your main browser is accessed.

3. Knowledge Base & Self-Improvement

Antigravity treats learning as a core primitive:

• Agents store useful context and code snippets in a knowledge base
• Past insights improve future task execution
• Project-specific patterns and preferences are retained
• Context is carried over across sessions for continuous improvement

4. Multi-Model Support

Although deeply optimized for Gemini 3, Antigravity offers model flexibility:

5. MCP (Model Context Protocol) Integration

Added on December 18, 2025, MCP support bridges the gap between the editor and your external tools:

• 30+ pre-built MCP servers available in the MCP Store
• One-click installation for popular services
• Custom MCP server configuration via mcp_config.json

Available MCP servers include:

• Databases: AlloyDB, BigQuery, Spanner, Cloud SQL, Supabase, MongoDB, Neon, Redis
• Dev-Tools: GitHub, Linear, Heroku, Netlify, Vercel, Firebase
• Productivity: Notion, Figma, Atlassian, Looker
• AI/ML: Perplexity, Pinecone, Prisma

---

Interaction Modes & Autonomy Levels

Development Mode Presets

When you configure Antigravity for the first time, you choose one of four preset modes that combine terminal execution and verification policies:

Planning Mode vs Fast Mode

Planning Mode (Default)

• Generates detailed implementation plans and task lists
• Creates step-by-step walkthroughs
• Offers opportunity for intervention at each step
• Ideal for complex projects that require oversight

Fast Mode

• Direct command execution without intermediate artifacts
• Immediate code changes and rapid iterations
• Best for quick fixes without major impact
• No intermediate verification required

Verification Preferences

Customize how much oversight you want:

• Never - Full autonomy, no interruptions
• Agent decides - AI determines when verification is needed
• Always - Human approval required at every step

---

Customization: Rules & Workflows

Antigravity offers powerful customization through rules and workflows, accessible via the ... menu → Customizations.

Rules

Rules guide the agent's behavior as system-level instructions. They are applied automatically without user intervention.

Example Rules:

Always use TypeScript for new files
Follow PEP 8 style guide for Python code
Document all public methods with JSDoc comments
Use async/await instead of callbacks
Never commit console.log statements

Storage Locations:

• Global: Applied across all Workspaces
• Per Workspace: Stored in the .antigravity/ directory of the project

Workflows

Workflows are saved prompts that are triggered on demand with / slash commands. Think of them as reusable prompt templates.

Example Workflow: Generate Unit Tests

name: generate-unit-tests
description: Generate comprehensive unit tests for the current module
prompt: |

  Generate unit tests for the current file using pytest.
  Include edge cases, error handling, and mock external dependencies.
  Follow AAA pattern (Arrange, Act, Assert).

Using Workflows:

1. Type / in the chat
2. Select from available workflows (e.g., /generate-unit-tests)
3. Agent executes with the predefined instructions

---

Agent Security Configuration

Giving an AI Agent access to your terminal and browser is a double-edged sword. Antigravity addresses this through a granular permissions system.

Terminal Command Auto-Execution Policies

Access via: Antigravity → Settings → Advanced Settings → Agent → Terminal

Allow Lists

Explicitly allow safe commands when the Terminal policy is set to Off:

{
  "allowedCommands": [
    "npm install",
    "npm run dev",
    "git status",
    "pytest",
    "eslint"
  ]
}

Deny Lists

Block dangerous commands even in Turbo or Auto mode:

{
  "deniedCommands": [
    "rm -rf",
    "sudo",
    "curl | bash",
    "chmod 777",
    "DROP TABLE"
  ]
}

Browser URL Allowlist

Limit which pages the Agent can visit to prevent prompt injection attacks:

Access via: Antigravity → Settings → Advanced Settings → Browser

{
  "allowedUrls": [
    "localhost:*",
    "*.google.com",
    "github.com",
    "stackoverflow.com"
  ]
}

Only domains in this list are accessible. All others are blocked.

Secure Mode (v1.11.17+)

Introduced on December 8, 2025, Secure Mode provides maximum oversight:

• Prevents Agents from autonomously executing targeted exploits
• Requires human review for ALL Agent actions
• Enforces security-conscious settings throughout the IDE

---

Changelog: December 2025 Updates

v1.13.3 (December 19, 2025)

Google Workspace Support

• Higher, more frequently updated rate limits for Google Workspace AI Ultra for Business subscribers
• Enterprise-ready features for team deployments

v1.12.4 (December 17, 2025)

Gemini 3 Flash Support

• Added support for Gemini 3 Flash model
• Faster execution for speed-optimized workflows
• 3 Improvements, 1 Patch

v1.11.17 (December 8, 2025)

Secure Mode & Security Fixes

• New Secure Mode option that:
  • Prevents agents from autonomously executing targeted exploits
  • Requires human verification for all agent actions
  • Enforces security-conscious settings
• Various security vulnerability patches (address Mindgard's reported backdoor concerns)

v1.11.14 (December 4, 2025)

Google One Support

• Higher, more frequently updated rate limits for Google AI Pro and Ultra subscribers
• 5-hour refresh cycle for longer coding sessions
• Weekly rate limits for free users (minimizes hitting limits during projects)

v1.11.9 (November 26, 2025)

Stability and Bugfixes

• Authentication flow bugfixes
• Improved stability platform-wide

v1.11.5 (November 20, 2025)

Nano Banana Pro Integration

• Enhanced image generation capabilities, anchored in your codebase
• Generate UI mockups, system diagrams, and embeddable assets
• Visual artifact creation from natural language prompts

---

Rate Limits & Pricing

Pricing Tiers

Free Tier

• Access to Gemini 3 Pro for all users
• Unlimited Tab Code Completions
• Full access to all product features
• Weekly-based rate limits (enables focused usage)
• Rate limits based on "work done" by agents (complex tasks consume more quota)

Google AI Pro Subscribers

• Priority access to Antigravity
• Higher rate limits
• 5-hour refresh cycle

Google AI Ultra Subscribers

• Highest, most generous rate limits
• Priority access during high demand
• All premium features

Google Workspace AI Ultra for Business

• Enterprise-Grade rate limits
• Team collaboration features
• Organizational controls

---

Known Issues & Bugs

As a public preview product, Antigravity has several known issues that early adopters should be aware of:

Critical Issues

Workflow Limitations

Locked-Down Planning Documents

• Planning documents, tasks, and walkthroughs are stored in a proprietary brain/ directory
• Not stored in your project folder
• Makes IDE switching difficult without manual copy/pasting

Extension Marketplace Issues

• Popular extensions (like OpenAI Codex IDE) hard to find in search
• Some users cannot install extensions
• Workaround: Import profile from Cursor or VS Code; scroll to the end of search results

Project Directory Issues

• Moving an Antigravity project directory silently breaks some functionality
• Knowledge Items may not persist after directory moves
• No obvious fix available

Common Workarounds

1. "Continue" Magic: If the agent pauses or throws vague errors, typing continue often resumes the task
2. Profile Import: Import your existing VS Code/Cursor profile to restore extensions
3. Scroll Search Results: Missing extensions may appear at the end of the Marketplace search

---

Security Vulnerabilities

⚠️ Warning: Antigravity's terms of service state that it "is known to have certain security limitations." Exercise caution with sensitive code.

Mindgard: Persistent Code Execution Vulnerability

Discovered: November 25, 2025 (within 24 hours of launch)

Problem: A malicious "trusted workspace" can embed a persistent backdoor that:

• Executes arbitrary code on every future application launch
• Also triggers even when no specific project is open
• Persists after complete uninstallation and reinstallation

Impact: A compromised workspace can affect all subsequent Antigravity usage regardless of trust settings.

Status: Google's Bug Hunters Team is evaluating; not yet deemed severe enough for reward qualification.

PromptArmor: AI Prompt Injection Risks

Findings:

• Default settings allow AI agents to automatically execute commands
• Untrusted inputs in source files can manipulate agents to execute unintended commands
• Agents can read sensitive files and generate content without strict enforcement

Identified Attack Vectors:

• Data exfiltration via Markdown, tool calls, or hidden instructions
• Prompt injection attacks via malicious code comments
• Unauthorized command execution via crafted source files

Recommended Security Practices

1. Disable "Auto-execute suggested terminal commands"
2. Turn off agent access to production credentials and deploy keys
3. Enforce strict source file sanitization and code review policies
4. Run security scans on both human-written and AI-generated code
5. Use sandboxed environments for sensitive projects
6. Enable Secure Mode (v1.11.17+) for maximum oversight
7. Configure browser allowlists to restrict agent browsing

Privacy Concerns

Users have raised questions regarding:

• Whether private code is used to train Google's models
• Closed-source nature raises long-term viability concerns
• All code is processed on Google's servers (no local inference option yet)

"If your code legally needs to stay local, Antigravity is not your tool." — Developer Community Feedback

User Feedback & Reception

Product Hunt Reviews

Rating: 4.6/5 (58 ratings, as of December 2025)

Positive Feedback:

• Implementation plans and walkthroughs for collaborative UX praised
• Agent "thought process" highly discoverable and transparent
• Workflow improvements over Cursor's linear chat
• Paying subscribers report not reaching limits

Critical Feedback:

• UI quirks and bugs (panels obscure Accept/Reject buttons)
• Allowlist for commands does not work for some users
• Credits/quota depletion forces work stoppage
• Some skepticism that it is a "VS Code Wrapper with a new skin"

Frequent Praise

"I'm very impressed with it so far. It's already helped me update my blog to the latest Hugo... it even recognized that some shortcodes from the old version changed and automatically updated my blog posts." — Mete Atamel, Google Developer Expert

• Multi-Agent Parallel Workflows praised
• Browser integration seen as a key differentiator
• Agent verification through artifacts builds trust
• Model-agnostic approach appreciated

Frequent Criticism

• Stability Issues: Frequent "model overloaded" errors during peak usage
• Quota Bottlenecks: Even paying users reach limits during intensive sessions
• "Astroturfing" Concerns: Some users suspect overly enthusiastic early reviews
• Junior Developer Analogy: "Incredibly fast, occasionally reckless, and needs supervision"

Community Verdict

"Right now, Antigravity feels like hiring a talented but inexperienced junior developer... The era of agentic coding is here, and even with the bugs, it's impossible to look away." — PromptLayer

Performance Benchmarks

Industry Standard Benchmarks

What These Mean

• SWE-bench Verified: Measures the ability to solve real-world GitHub issues in production codebases
• Terminal-Bench 2.0: Tests tool usage capabilities, chaining multi-stage workflows via Terminal
• WebDev Arena: Evaluates Web UI generation quality

Practical Performance

Strengths:

• Complex multi-stage workflows
• Large codebase understanding (1M+ Token context window)
• Cross-File Refactoring
• End-to-End UI and Backend work

Weaknesses:

• High latency during peak load
• Occasional task abandonment
• Code duplication bugs (Agent can duplicate code sections)

---

Security Features

Agent Security Controls

• Allow Lists: Specify which pages/resources agents can access
• Deny Lists: Block access to sensitive systems
• Browser Allow Lists: Control which websites the browser agent can visit

Secure Mode

Introduced in v1.11.17, this mode offers maximum oversight:

• Human review required for all agent actions
• Prevents autonomous exploit execution
• Enterprise-compliance ready

Isolated Browser Environment

• Separate Chrome profile from personal browsing
• No access to main browser cookies, history, or accounts
• Specialized extension for secure control

---

Getting Started

Installation

macOS (Homebrew):

brew install --cask antigravity

Linux (apt):

sudo apt install antigravity

Direct Download:

Visit antigravity.google/download

First Launch

1. Run the installer for your OS
2. Answer the setup questions and click Next
3. Select "Start fresh" for a new Workspace
4. Choose your autonomy preferences
5. Sign in with your Google Account
6. Install the Chrome extension when prompted

Quickstart Workflow

1. Create a new Workspace
2. Describe your project goal in natural language
3. Review the generated implementation plan
4. Approve or modify the task list
5. Let the agent execute while you monitor artifacts
6. Provide feedback directly on each artifact
7. Push to production when satisfied

---

Tips for Maximum Productivity

Best Practices for Getting Started

1. Start with Agent-Assisted Development - Learn how agents work before relinquishing control
2. Use Planning Mode for complex features - Review implementation plans before execution
3. Utilize MCP Server for your tech stack (Context7 for Docs, GitHub for Repos)
4. Remove Testing Tasks from auto-generated plans if you prefer manual testing
5. Enable Agent Auto Fix Lints for automatic syntax and formatting corrections

Workflow Optimization

6. Use the Playground for experiments before committing to a workspace
7. Save Summaries as context for multi-feature projects
8. Regularly Review Artifacts - Leave comments to guide agent behavior
9. Create Separate Folders for different tasks to avoid context confusion
10. Write Focused, Goal-Oriented Prompts - Avoid mixing unrelated tasks

Troubleshooting

11. Type "continue" when the agent pauses or throws vague errors - often resumes the task
12. Break Complex Tasks into Smaller Parts if agents terminate mid-execution
13. Use Review-Driven Mode for critical paths (payment systems, authentication)
14. Monitor Quota Notifications and plan accordingly - Utilize non-AI tasks during cooldown periods

Strategies for the Free Tier

15. Reserve Complex Projects for when limits are refreshed
16. Set Aside Simple Tasks (code snippets, small bug fixes) for the free tier
17. Utilize Focused Sessions - Weekly limits allow burst usage on projects

Security-First Approach

18. Configure Browser Allowlists to restrict agent browsing
19. Enable Secure Mode for maximum oversight
20. Review All Agent Actions before approving in sensitive codebases

Antigravity vs. Competition

Key Differentiators

Antigravity's Advantages:

• Only IDE with native browser automation for testing
• Multi-Agent parallel orchestration via Manager View
• Verifiable artifacts for transparent AI work
• Completely free during the preview
• Massive 1M+ Token context window

Antigravity's Disadvantages:

• Preview product with stability issues
• Security concerns are still being addressed
• All code is processed on Google's servers
• Locked-in planning documents (Vendor Lock-in Risk)

---

What's Next

Based on Google's roadmap and community discussions:

Confirmed / In Progress

Expected Improvements

• WebGPU-accelerated Local Inference - Handle privacy-sensitive code without the cloud
• Live Collaborative Editing - Code simultaneously with team members and agents
• Expanded MCP Ecosystem - More pre-built integrations
• Improved Stability - Address "model overload" issues
• Security Hardening - Fix reported vulnerabilities

Industry Implications

"The teams that will dominate in 2026 aren't waiting for perfection. They're learning on imperfect tools now."

The shift from "Tipper" to "Architect" is accelerating. Developers who can architect agent workflows will become increasingly valuable, while boilerplate-focused roles face disruption.

---

Conclusion

Google Antigravity represents a bold bet on the future of software development.

By giving agents dedicated space to work autonomously – across editor, terminal, and browser – while offering humans transparent artifacts for verification, it addresses the trust gap that has constrained earlier AI coding tools.

The Good

• Agent-First Architecture that treats AI as a first-class citizen
• Verifiable Artifacts that build trust through transparency
• Native Browser Integration for end-to-end testing
• Multi-Model Support that avoids vendor lock-in on AI models
• Free During Preview with generous limits

The Concerns

• Preview Stage Stability with frequent bugs and crashes
• Security Vulnerabilities are still being addressed
• All Code is Processed on Google's Servers (no local option)
• Vendor Lock-in Risk with locked-in planning documents
• Steep Learning Curve for Agent-First Workflows

The Verdict

"Right now, Antigravity feels like hiring a talented but inexperienced junior developer: incredibly fast, occasionally reckless, and in need of supervision. Download the preview to experience the shift from 'Tipper' to 'Architect' firsthand, but keep your version control tight."

If you're ready to embrace the agentic future with all its current imperfections, Antigravity offers a compelling glimpse into where development is headed.

For production-critical work, consider starting with Agent-Assisted or Review-Driven mode until the platform matures.

Download Antigravity today at antigravity.google/download and experience the liftoff – but maybe keep a backup parachute handy.

---

Last updated: December 27, 2025

Sources: Google Developers Blog, Releasebot, InfoWorld, VentureBeat, Google Codelabs, Mindgard Security Research, PromptArmor, Product Hunt, Community Tutorials and User Feedback