Antigravity 3.0 (New Upgrades): These New Updates make ANTIGRAVITY REALLY GOOD!

Anti-Gravity Updates: Skills, Rate Limits, and Secure Mode

Key Concepts:

• Anti-Gravity: Google’s AI-powered coding assistant.
• Skills: Saved workflows or custom agents triggered on demand via slash commands.
• Rules: System instructions the agent always follows.
• Rate Limits: Usage quotas for AI models, now weekly for most users.
• Secure Mode: A set of security measures to control agent actions, including terminal permission modes and allow/deny lists.
• Thinking Tokens: Internal reasoning processes by Gemini 3 models that contribute to usage quota consumption.
• Opus 4.5, Gemini 3 Pro, Gemini 3 Flash: Specific AI models available within Anti-Gravity.
• Google AI Pro/Ultra: Subscription tiers offering different levels of access and quotas.

1. Introduction & Previous Criticism

The video details recent updates to Google’s Anti-Gravity, an AI coding assistant. The speaker acknowledges previous criticisms regarding the tool’s buggy agent manager and confusing rate limits, noting that Google has addressed many of these issues based on user feedback. The core updates discussed are the introduction of “Skills,” changes to the rate limit structure, and the implementation of “Secure Mode.”

2. Skills: Custom Workflows on Demand

A significant addition is the “Skills” feature. These function as saved workflows or custom agents, similar to Claude Code Commands or Ruse custom commands, allowing users to trigger specific agent behaviors for defined tasks. Skills can be created globally or per workspace and activated using slash commands.

• Distinction from Rules: Skills are triggered on demand by the user, unlike “Rules” which are always active system instructions (e.g., always adding docstrings).
• Use Cases: Skills are applicable to tasks like code reviews, documentation generation, test writing, and deployment workflows.
• Team Collaboration: Skills can be shared across teams, ensuring consistent results and workflows.
• Connection to Previous Work: The speaker draws a parallel to custom commands demonstrated in a previous video on a “self-spawning AI coder,” highlighting that Skills are now natively integrated into Anti-Gravity.

3. Rate Limit Updates: A Mixed Bag

Google has implemented weekly rate limits for all models, including for Google AI Pro users. This change aims to balance quota allocation and fairness due to high demand.

• Shift from Hourly to Weekly Limits: Previously, limits refreshed every 5 hours, considered generous. The new weekly limits can be restrictive for heavy users who exhaust their quota early in the week.
• Tiered Impact: The limits primarily affect Google AI Pro users. Google AI Ultra subscribers ($200/month) remain unaffected and are positioned as the best option for “Power developers.”
• Model Access: Despite the limits, Pro users retain access to models like Opus 4.5, Gemini 3 Pro, and Gemini 3 Flash.
• Usage Calculation: Anti-Gravity calculates usage based on the work done by the agent, not simply the number of requests. Complex reasoning tasks consume more quota due to “thinking tokens” generated during the Gemini 3 model’s internal reasoning process. This means planning and debugging will use quota faster.
• Continued Access: All users, regardless of tier, maintain unlimited tab completions and access to the agent manager and browser integration.

4. Secure Mode: Enhanced Security Measures

“Secure Mode” addresses previous security concerns identified shortly after Anti-Gravity’s launch, where vulnerabilities could potentially lead to data leaks or unauthorized command execution. Google’s terms of use acknowledge inherent security limitations.

• Terminal Permission Modes:
  • Turbo Mode: Automatic execution of all terminal commands (fastest, riskiest).
  • Auto Mode: Agent analyzes commands and requests approval when necessary (balanced speed and oversight).
  • Off Mode: Explicit human approval required for every command (slowest, safest – recommended for production branches).
• Allow/Deny Lists:
  • Allow List (Positive Security Model): Only explicitly allowed commands are permitted. Considered the most secure.
  • Deny List (Negative Security Model): All commands are allowed except those specifically blocked. Useful for speed with targeted restrictions.
• Enterprise Recommendations: Google recommends running Anti-Gravity within a sandbox or VM, utilizing command allow lists, and employing version control for auditing changes.

5. Value Proposition & Overall Assessment

The speaker maintains that Anti-Gravity remains a strong option despite the rate limit changes.

• Free Tier: The free tier is considered the best available for an AI editor, offering access to various models.
• Pro Tier Value: While less generous than before, the Pro tier still provides access to valuable models like Opus 4.5 and includes 2TB of Google Drive storage, Gemini Pro for mobile, and improved Gemini CLI/Jewel limits.
• Agent Manager Caveats: The agent manager remains somewhat buggy, and the side panel agent is still recommended for stability.
• Quote: “The free tier is still the best free tier that you can get anywhere for an AI editor.”

6. Conclusion

The updates to Anti-Gravity, particularly Skills and Secure Mode, are viewed positively, enhancing customization and security. The rate limit changes are a drawback for Pro users, but the tool remains usable and offers good value, especially considering the free tier. The speaker encourages viewers to try Anti-Gravity, starting with the free tier, and to consider the Ultra tier for unlimited access if needed.