Google is folding Antigravity, its agentic coding system, directly into Gemini Enterprise Standard and Plus subscriptions, eliminating the need for separate licenses and putting AI agent oversight into the hands of IT administrators. The shift signals a broader move toward treating powerful AI agents as enterprise infrastructure rather than standalone developer tools, complete with workspace sandboxing, budget controls, and centralized audit logging.

Why Does Enterprise Control Over AI Agents Matter?

Antigravity is not a simple code-completion tool. Google's agents can inspect entire codebases, execute terminal commands, operate web browsers, and carry out multi-step tasks autonomously. That power creates a governance challenge: an agent that can run code locally or control a browser needs firm boundaries to prevent unintended consequences. By bundling Antigravity into Gemini Enterprise, Google is addressing that tension head-on, giving organizations a single control plane for agent access, spending, and accountability.

The administrative layer is the real innovation here. Rather than developers acquiring Antigravity as a separate product, IT and engineering leaders can now enable it as part of their existing subscription, using the same identity systems, budget frameworks, and audit trails they already manage. This reduces friction for organizations that want to deploy agentic coding at scale without fragmenting their security and compliance infrastructure.

What Controls Do Administrators Actually Get?

Google says Gemini Enterprise administrators can configure several layers of oversight:

• Workspace Sandboxing: Isolate agent activity to specific project environments and restrict browser or Model Context Protocol server permissions based on organizational policy.
• Budget and Quota Management: Set pooled quotas, budget caps, and capped overages across projects, teams, and individual developers to prevent runaway costs.
• Centralized Audit Logging: Track all prompts, agent tool executions, and generated artifacts in a single log, enabling compliance reviews and incident investigation.
• Identity Integration: Use Workforce Identity Federation and Application Default Credentials to authenticate agents through existing enterprise systems, eliminating the need for developers to manually configure API keys.

Google also states that enterprise activity remains within the customer's enterprise boundary under Google Cloud Terms of Service, customer code is not used to train its base models, and existing Identity and Access Management (IAM) policies and VPC Service Controls are respected. These are product commitments from Google, not independent security audits, but they address the core concerns enterprises raise about deploying AI agents in regulated environments.

How to Set Up Antigravity Across Your Development Team

For organizations adopting Antigravity through Gemini Enterprise, the rollout involves several practical steps:

• Configure Rules Files: Create markdown-based constraint files that tell the agent what it should and should not do in your specific environment. Google provides two scopes: global rules in a.gemini folder in the home directory, and workspace-specific rules in a.agents/rules folder at your project root. These rules can be always active, triggered by the model when relevant, invoked manually, or applied to specific file types.
• Define Agent Customization Surfaces: Google's customization panel includes six surfaces: Model Context Protocol (MCP) servers, Skills, Rules, Plugins, Hooks, and Sidecars. Start with Rules to establish guardrails, then layer in other customizations as your team's needs evolve.
• Test Across Your Editor Stack: Antigravity extensions are now available for VS Code, JetBrains IDEs, Visual Studio 2026, and Zed. A single Antigravity account shares session context across supported editors, so developers can switch environments without losing agent context or starting a new session.

The practical value of this approach became clear when one user set up Antigravity according to Google's official documentation. By creating a rules file that described their project structure, naming conventions, and constraints, the agent stopped making irrelevant suggestions and began providing contextually appropriate guidance. "The agent wouldn't know my folder layout or design conventions so it defaulted to assumptions that didn't apply to my projects," the user explained. Once rules were in place, the agent retained that context across sessions.

What's Still in Preview, and Why It Matters?

The rollout is not complete. VS Code extensions are available on macOS, Linux, and Windows, but Visual Studio 2026 support for.NET solutions remains in preview. Enterprise support for JetBrains and Zed also remains in preview, meaning organizations with mixed development stacks may encounter inconsistencies in how identity, context, and controls work across different editors.

This is the critical test for Google's strategy. The company's differentiator is not simply that Antigravity can appear in multiple editors; it is that the same identity, session context, and administrative policies should work consistently across VS Code, JetBrains, Zed, and Visual Studio. If that continuity breaks down in preview-supported environments, the promise of a unified enterprise control plane falls apart. Organizations with developers using several editors will need to monitor rollout quality closely before committing to Antigravity as an organizational layer.

The shift from standalone product to enterprise entitlement represents a maturation of agentic coding. Google is betting that enterprises will adopt AI agents more readily if they can govern them the same way they govern other critical infrastructure. Whether that bet pays off depends on whether the promised continuity of account, context, and controls actually holds up in real mixed-editor organizations.