Antigravity 2.0 & Gemini 3.5 Flash (Fully Tested): SO BAD! FINAL NAIL IN THE COFFIN FOR GOOGLE.

Key Concepts

• Anti-gravity 2.0: Google’s new AI agent ecosystem for coding, featuring a CLI, SDK, IDE, and standalone desktop application.
• Gemini 3.5 Flash: A new, faster iteration of Google’s model, noted for significant price increases.
• Gemini Spark: A new feature integrated into the Gemini desktop app, currently limited to trusted testers and high-tier subscribers.
• Sub-agents: AI agents spawned by a primary agent to handle specific tasks, a feature now integrated into the Anti-gravity framework.
• Electron Apps: The underlying framework for the Anti-gravity desktop app, criticized for poor UI/UX implementation.

---

1. Overview of Google I/O Announcements

Google recently unveiled several updates, with the most significant for developers being Anti-gravity 2.0, Gemini 3.5 Flash, and Gemini Spark. The presenter notes that the presentation demos were basic and lacked depth. While Gemini 3.5 Flash boasts speeds 12 times faster than previous iterations, it comes with a steep price hike: $1.50 for input and $9.00 for output per million tokens, placing it in the same pricing tier as high-end models like GPT-5.3.

2. The Anti-gravity Ecosystem

Anti-gravity 2.0 is structured into four distinct interfaces:

• CLI (Command Line Interface): Described as buggy and unreliable, with frequent authentication failures. The presenter advises against using it, suggesting alternatives like OpenCode.
• SDK: Allows developers to harness Gemini models programmatically.
• IDE: A VS Code fork with an integrated AI agent.
• Anti-gravity App: A standalone desktop application for non-coders or those who prefer a GUI.

3. Critical Analysis of the Anti-gravity Desktop App

The presenter highlights several major flaws in the Anti-gravity desktop application:

• Lack of Innovation: The interface is described as a direct copy of the "Codex" desktop app, which itself was inspired by tools like Conductor.
• UI/UX Bugs: The app suffers from resizing issues (sidebar disappearing), lack of terminal access, and an inability to open folders directly from the interface.
• Navigation Issues: The "change files" view fails to scroll automatically to the relevant sections, forcing manual user intervention.
• Developer Tools: The app allows users to open developer tools with a single click, which the presenter views as a sign of an unfinished, unpolished product.

4. Performance and Benchmarking

• Cursor Bench: In third-party benchmarks, the new Gemini models underperform compared to competitors like Opus and GPT.
• Simulation Testing: The model struggled with complex simulations (e.g., folding tables, contact lenses), producing "wonky" animations and camera views.
• SVG Focus: The presenter criticizes Google for over-optimizing the model for SVG generation rather than functional, real-world coding tasks.
• Elevator Simulator: While the model successfully created a functional elevator simulator, the aesthetic quality was inferior to outputs generated by Claude.

5. Recommendations and Alternatives

The presenter concludes that the current Anti-gravity offering is not worth the $200 subscription fee. Key takeaways include:

• Better Alternatives: The presenter recommends GLM 5.1 or Kimmy, noting that GLM currently offers a 50% discount for migrating users.
• Verdict: Unless used for free, the Anti-gravity suite is deemed "atrociously bad" and uncompetitive compared to existing tools like Verdant or T3 Code.
• Final Thought: Google is currently in a "catch-up" phase rather than an innovation phase, and the current state of their AI coding tools is not recommended for professional workflows.

"I do think that Google should innovate and not catch up if it makes sense. Anti-gravity is so unfinished that you can literally with one click open the whole developer tools and inspect the code." — Presenter