Google Upgrades Antigravity to Tackle Genomics and Multi-Day Coding Projects
Google's agentic IDE ships a bundle of upgrades spanning genomics skills, deep-reasoning multi-agent workflows, generative UI artifacts, and a rebuilt terminal.
- AlphaGenome Atlas Skills plug the 1PB variant database into Antigravity for agent-driven genomics workflows.
- New
/boostslash command runs a three-tier multi-agent reasoning pipeline for hard bugs and refactors. - /teamwork-preview is now open to all paid plans, not just $200 Ultra.
- Generative UI artifacts render interactive dashboards and 3D simulations inline, exportable as standalone HTML.
- Rebuilt embedded terminal and Git integration let agents run tests and commands without leaving the IDE.
- AlphaGenome remains research-only and is not approved for clinical or diagnostic use.
Google has expanded Antigravity, its agentic integrated development environment, with genomics research skills, deeper multi-agent reasoning, interactive visual artifacts, rebuilt Git controls, and an embedded terminal. The additions let agents plan and execute more work inside the IDE, from prioritizing genetic variants to coordinating multi-day software projects.
AlphaGenome brings variant triage into the IDE
The new AlphaGenome and AlphaGenome Atlas Skills connect Antigravity’s assistant to the one-petabyte AlphaGenome Atlas. The database contains precomputed predictions for 9 billion single-nucleotide variants, which are single-letter changes in human DNA. Researchers can query those predictions without running the underlying machine-learning model for every variant.
Inside Antigravity, the skills can analyze large variant sets, rank likely effects, explain the biological signals behind each ranking, and generate plots without requiring researchers to write analysis code. Supported tasks include:
- Variant prioritization: Rank large lists of uncharacterized variants with the AlphaGenome Variant Impact (AVI) score.
- Effect explanations: Describe why a variant ranks highly and identify predicted disruptions such as splicing defects.
- Visual analysis: Create reference-versus-alternate allele plots inside the chat.
- Atlas navigation: Generate parameterized links that open the relevant tracks on the Atlas website.
AlphaGenome is intended solely for research, hypothesis generation, and variant prioritization. It lacks validation and regulatory approval for clinical diagnosis.
Three execution modes for different workloads
Antigravity now offers three levels of agent execution. Default mode begins work from a single prompt, the new /boost command applies deeper reasoning without a scoping interview, and /teamwork-preview scopes the assignment before coordinating a larger agent team.
| Mode | How it starts | Best suited to |
|---|---|---|
| Default | Executes a single prompt immediately | Feature work and interactive pair programming |
/boost |
Executes immediately with deeper multi-agent reasoning | Concurrency bugs, algorithms, and substantial refactoring |
/teamwork-preview |
Runs a two-phase scoping interview before execution | Large, open-ended, or multi-day projects |
When invoked, /boost creates a three-tier hierarchy consisting of a top-level Orchestrator, DeepCoder or DeepInvestigator coordinators, and isolated execution workers. Independent verification loops check the resulting work. Sessions can last from seconds to hours, and the mode consumes more tokens than standard execution.
/boost is available to Pro and Ultra subscribers in Antigravity 2.0 and the Antigravity CLI.
Teamwork separates roles from orchestration
/teamwork-preview targets software projects, simulations, and research assignments that span many files, specialties, or work sessions. It coordinates specialized agents and verification steps across the broader assignment.
The implementation separates orchestration logic from agent descriptions. Each teamwork pattern declares the required roles and behaviors, and the framework interprets that specification to create and coordinate the agents. Reusable mechanisms such as adversarial critique loops can therefore operate across software, mathematics, simulation, and research tasks.
/teamwork-preview, previously restricted to the $200-per-month Ultra plan, is now available on all paid plans. Google says it has used the system internally for mathematics, hardware simulation, and open-source optimization.
Interactive artifacts stay local
Antigravity’s Generative UI system lets agents create interactive artifacts inside the Agent Manager or a dedicated artifact panel. Outputs can range from dependency graphs to three-dimensional simulations, giving developers an inspectable interface for results that are difficult to evaluate as text.
- Create: Invoke
/generative_uito request a visualization. - Inspect: Click elements in the rendered preview to examine them.
- Run locally: Use zero-dependency artifacts without downloading additional packages, including while offline.
- Export: Save the result as a standalone HTML file for later use.
Git and the terminal stay in the loop
The release also adds a rebuilt Git version-control interface and an embedded terminal. Developers can track repository changes and run tests, scripts, and shell commands within Antigravity, keeping code generation, inspection, and verification in the same workspace.
With these additions, Antigravity combines routine coding assistance, deeper reasoning for bounded problems, coordinated agents for long-running projects, inspectable visual output, local development tools, and a domain-specific genomics interface. Planning, execution, verification, and output can now remain within a single environment.