Google's Antigravity agent IDE arrives as a VS Code extension

Google's agent-first IDE Antigravity no longer requires developers to switch editors. According to dev.to, Google has shipped Antigravity extensions for VS Code, Visual Studio, JetBrains IDEs and Zed, bringing the same agent, review workflow and account into tools developers already use. The original Antigravity desktop application, which launched in November 2025, offered full agentic capability but only inside its own dedicated app.

One agent, three entry points

The dev.to walkthrough describes three ways into the same system: the full Antigravity 2.0 desktop app (itself a fork of VS Code), the agy command-line tool, and the new IDE extensions. All three sit on a local agy backend service that the extension installs automatically on first launch, feeding an Antigravity Agent Runtime that handles planning, terminal execution and a browser subagent, powered by Gemini.

The extension itself is deliberately thin: a side panel, an inline diff viewer and a renderer for the agent's plans and artifacts, layered on top of an existing VS Code install. Because the backend is shared, sign-in is unified across every surface — authenticate once and the same entitlements, permissions and conversation history follow whether you work from VS Code, JetBrains or a terminal.

Setup and access

Running the extension requires VS Code 1.90 or later on macOS, Linux or Windows. Installation is a matter of searching for Google Antigravity in the marketplace and signing in with a Google Account from the Activity Bar; the backend installs itself in the background. A paid plan is not needed — dev.to notes the free tier is enough to start. Enterprise teams can instead authenticate through Gemini Enterprise–enabled accounts, which ties usage to organization-level budgets and IAM policies rather than personal quotas.

Plans, walkthroughs and a browser subagent

Day-to-day work happens in a side panel where users start conversations, attach images, switch agent modes and choose models without leaving their code. A toolbar tracks open file changes, running terminal processes and artifacts as a task executes.

For non-trivial requests, the agent does not edit files immediately. It first produces an implementation plan — an artifact describing what it intends to change and why — and pauses for sign-off unless the review policy is set to always proceed. Developers can comment directly on the plan, hit Proceed, or submit feedback for the agent to revise against; the dev.to post compares this to reviewing a pull request description before the pull request exists. After a task finishes, a walkthrough artifact summarizes what actually changed.

For web work specifically, Antigravity includes a browser subagent that can open, read and actuate a real local Chrome instance: launching a dev server, clicking through a UI it just built, capturing screenshots and saving the interaction as a video artifact.

Permissions: deny beats ask beats allow

Every sensitive action — reading a file, running a command, hitting a URL, calling an MCP tool — is modeled as a permission resource in the form action(target) and evaluated against three lists. Deny blocks immediately, Ask pauses for explicit approval, and Allow runs without prompting. Precedence is strict: if a command matches both an Allow and a Deny rule, Deny wins, and anything matching no rule falls back to prompting the user. That structure is designed for broad allow rules combined with explicit denies for the few commands you never want executed automatically.

Why it matters

Agentic coding tools have largely asked developers to abandon their carefully tuned editor setups. By shipping the same backend as an extension alongside the desktop app and CLI, Google removes that friction and puts its Gemini-powered agent in direct competition with the AI assistants already living inside VS Code. The plan-first review flow and the explicit permission engine address the trust question that comes with an agent that can read files, run commands and drive a browser. With a free entry tier and enterprise hooks for governed deployment, the story here is portability: the agent becomes a layer over your existing setup rather than an application you have to move into.