Antigravity 3.0 (New Upgrades): Gemini 3.6, Agent Team-Up Option, Workbench, Low Thinking Effort

Key Concepts

• Antigravity: An AI-powered agent platform for coding and development.
• Teamwork Preview: A multi-agent orchestration feature that runs specialized sub-agents in parallel.
• Science Skills Bundle: A collection of tools and databases integrated into Antigravity to facilitate scientific research workflows.
• Gemini 3.5 Flash: The primary model used within Antigravity, now updated for better endurance and efficiency.
• Agentic Platform: A system where AI agents perform complex, multi-step tasks rather than just responding to simple prompts.
• Token Efficiency: The practice of managing AI model usage to minimize costs and resource consumption.

---

1. Teamwork Preview Expansion

Google has expanded the availability of the {slash} teamwork preview feature from the $200 "AI Ultra" plan to all paid plans.

• Functionality: This feature enables multi-agent orchestration, where a team of specialized sub-agents (orchestrator, implementer, reviewer, stress-tester, auditor) works in parallel.
• Performance: It is designed for complex tasks, such as building functional operating systems. For example, one project required 93 sub-agents, 15,000+ model calls, and over 2.6 billion tokens.
• Strategic Use: Due to high token consumption, it is recommended for large-scale refactoring, full app builds, or tasks requiring independent verification rather than minor edits.

2. Google DeepMind Science Skills Bundle

This update transforms Antigravity into a "scientific workbench" by providing agents with specialized knowledge and tool-use capabilities.

• Capabilities: The bundle includes skills for genomics, structural biology, cheminformatics, and literature search.
• Integrated Databases: It connects to over 30 scientific resources, including AlphaFold Database, UniProt, PubMed, arXiv, ChEMBL, ClinVar, PDB, and OpenAlex.
• Grounding: The primary goal is to improve reliability and provide citations/references, ensuring that scientific claims are backed by specific database queries rather than generic AI hallucinations.
• Implementation: Users can access this via the "Build with Google" settings. Some tools require the UV package manager and specific API keys.

3. Gemini 3.5 Flash Updates

Two significant changes have been made to the model integration:

• Improved Endurance: A new version of Gemini 3.5 Flash has been released, specifically tuned to maintain structure and reliability during long, complex coding tasks.
• Gemini 3.5 Flash "Low": A new "low-effort" mode designed for simple, low-risk tasks like fixing typos, updating documentation, or minor CSS adjustments. This mode is optimized to save tokens and preserve user quotas.
• Rate Limit Reset: Google reset rate limits for all users to allow for immediate testing of the improved model performance.

4. CLI and Desktop Session Syncing

With the release of Antigravity CLI V1.0.4, users can now maintain continuity across different interfaces.

• Workflow: Users can start a session in the Antigravity desktop app and resume it in the CLI using the {slash} resume command.
• Benefit: This allows developers to switch between the graphical interface (for agent management and planning) and the terminal (for speed) without losing context or conversation history.

---

Synthesis and Conclusion

Google is rapidly evolving Antigravity from a standard AI code editor into a comprehensive, multi-surface agent platform. The shift toward multi-agent orchestration ({slash} teamwork) and domain-specific intelligence (Science Skills) indicates a focus on high-complexity, professional-grade workflows.

Actionable Takeaways:

• Cost Management: Users should adopt a tiered approach to model usage: "Flash Low" for simple edits, standard "Flash" for general coding, and "Teamwork" only for major architectural tasks.
• Reliability: The integration of scientific databases provides a framework for more grounded, verifiable research, though human oversight remains essential for sensitive scientific or medical data.
• Platform Integration: The session syncing between CLI and Desktop significantly improves developer experience, allowing for a more fluid transition between different development environments.