Google’s AI coding agent just escaped its own IDE
When Google launched Antigravity in November 2025, it was on the premise that developers could hand an entire coding task The post Google’s AI coding agent just escaped its own IDE appeared first on The New Stack .
Google's AI coding assistant, Antigravity, has begun integrating into developers' favorite code editors, expanding beyond its original desktop application. This move aims to make the tool more accessible to engineers already working within their preferred development environments. The company announced the new extensions for Visual Studio Code, Visual Studio, JetBrains IDEs, and Zed, as well as the availability of Antigravity through eligible Gemini Enterprise subscriptions.
These extensions enable developers to open conversations with the AI agent in a side panel, review changes inline, inspect plans, and delegate multi-step tasks without needing to leave their current workflow. The same Antigravity account functions across all supported environments, eliminating the need for separate sign-ins or license management.
Availability is currently offered for Visual Studio Code (macOS, Linux, and Windows), Visual Studio 2026 and .NET solutions in preview, JetBrains IDEs (IntelliJ IDEA, PyCharm, WebStorm, GoLand, CLion, and Rider), and Zed. Enterprise administrators can manage tool usage through Gemini Enterprise Standard, Plus, and Standard Emerging Market plans, with the option to set monthly spending limits for each Google Cloud project.
Once the allotted quota is exhausted, Antigravity either disables itself or shifts to pay-as-you-go pricing. Individual users and teams are metered using a rolling seven-day pool, with any unused credits disappearing when the pool resets. A single nontrivial engineering task may consume between 150,000 and 200,000 tokens, while multi-agent handoffs add more input tokens with each transfer.
Token spending can escalate rapidly, as evidenced by a built-in Claude Code skill recently found to load over 200,000 tokens before providing an answer. To address this concern, Google is not the first to implement token budgets for enterprise AI usage. Microsoft introduced AI token budgets for its engineering divisions after discovering significant token expenditure among its engineers.
Similarly, Uber exhausted its entire 2026 AI coding budget within the first four months of the year, and an internal Amazon project exceeded its planned budget by 860%. Security policies are enforced through authentication, with developers signing in using company credentials and selecting the specific Google Cloud project and region for Antigravity.
Organizations can integrate with existing identity providers via Workforce Identity Federation, while developers can authenticate using Application Default Credentials. Once authenticated, agent sessions inherit the organization's IAM policies, VPC Service Controls, and regional data boundaries, ensuring that data from enterprise Antigravity sessions remains separate from training activities for the foundation models.
However, connecting to a production system using Antigravity requires careful attention, as misconfigured settings can lead to serious consequences. Administrators can limit an agent's workspace, block browser access, and specify which MCP servers the agent can use. In cases where a connection reaches a production system, a poorly configured setting could have severe implications, similar to the incident where ElevenLabs' MCP server permitted Claude to delete production voice agents from a chat window.
By integrating Antigravity into existing editors and managing its usage through a control plane, Google aims to bring its AI coding agent to developers' machines without requiring widespread tooling changes, while maintaining consistent control and billing across all supported environments.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
