Google Antigravity Updates: Teamwork, Science Skills, and Model Tiers

Multi-Agent Orchestration and Research Workflows

Google has expanded access to the /teamwork preview feature, which was previously restricted to the $200 Google AI Ultra plan, to all paid Antigravity subscribers. This feature enables multi-agent orchestration where specialized sub-agents operate in parallel to handle complex coding tasks, such as building full applications or performing large-scale refactors. To utilize this, users invoke the /teamwork command, which distributes tasks across agents responsible for exploration, implementation, review, and stress testing.

Additionally, the platform introduced a "Science Skills" bundle developed by Google DeepMind. This plugin transforms the editor into a scientific workbench by providing agents with specific instructions and access to over 30 scientific databases and tools, including AlphaFold Database, UniProt, PubMed, and OpenAlex. The integration relies on the uv package manager for dependency handling and is designed to improve grounding and citation accuracy for research-heavy tasks in genomics, structural biology, and chemistry.

Model Optimization and Surface Syncing

Antigravity has updated its internal implementation of Gemini 3.5 Flash to improve endurance on complex, long-running tasks. Alongside this, a new "Gemini 3.5 Flash Low" mode is available for low-effort tasks such as documentation updates, simple UI tweaks, and minor code edits. This mode is intended to reduce token consumption for non-reasoning-heavy operations, allowing users to preserve their quotas for more demanding work.

Finally, the platform improved cross-surface usability with session syncing between the Antigravity desktop application and the AGY CLI (v1.0.4). Users can now initiate a session in the desktop interface and resume the same conversation thread within the terminal by using the /res command, facilitating a more fluid transition between graphical and command-line workflows.