Antigravity Turns the IDE into a Control Plane for Agentic Coding

Overview

Google introduced Antigravity as an agentic development platform built on top of Gemini 3. Launched on November 18, 2025, Antigravity is more than an autocomplete or pair programming assistant. It aims to reframe the IDE as a control plane where autonomous agents plan, execute, and explain complex software tasks across the editor, terminal, and browser.

What Antigravity Does

Antigravity behaves like a modern AI editor while treating agents as first class workers. Agents can decompose tasks, coordinate with other agents, edit files, run commands, and drive a browser. Developers interact at the task level and the platform manages low level tool interactions, allowing agents to act asynchronously and autonomously on end to end workflows.

Implementation and Availability

Under the hood Antigravity is an Electron application forked from Visual Studio Code. It requires signing in with a Google account and is available as a free public preview on macOS, Linux, and Windows.

Models, Pricing, and Runtime

The platform exposes multiple foundation models inside the same agent framework. In the preview, agents can use Gemini 3, Anthropic Claude Sonnet 4.5, and OpenAI GPT OSS models, giving developers model optionality in a single IDE. For individual users Antigravity is free. Google states Gemini 3 Pro usage is subject to generous rate limits that refresh every five hours and expects only a small fraction of power users to reach those limits.

Editor View and Manager View

Antigravity provides two primary work modes. Editor view is the default and resembles a standard IDE with an agent in a side panel. The agent can read and edit files, propose inline changes, and use the terminal and browser when needed. Manager view shifts focus from single files to multiple agents and workspaces, acting as a mission control for coordinating asynchronous agent runs.

Artifacts Instead of Raw Logs

A central design choice is the Artifact system. Rather than exposing raw tool call logs, agents emit human readable artifacts that summarize what they are doing and why. Artifacts are structured objects that can include task lists, implementation plans, walkthrough documents, screenshots, and browser recordings. They present work at the task level and are intended to be easier to verify than dense traces of model actions.

Google presents Artifacts as a response to trust issues in current agent frameworks. By surfacing task level artifacts plus verification signals, Antigravity aims to let developers audit agent work without being overwhelmed by internal step by step logs or left guessing at hidden behavior.

Design Tenets and Feedback

Antigravity is explicitly built around four tenets: trust, autonomy, feedback, and self improvement. Trust is supported through artifacts and verification steps. Autonomy comes from agents having access to editor, terminal, and browser surfaces, enabling more complex workflows without constant prompting. Feedback is provided via comments on artifacts, and self improvement relies on agents learning from past work and reusing successful procedures.

Developers can comment directly on specific artifacts, including text and screenshots. Agents can incorporate this feedback into ongoing work without restarting a run. The platform also includes a knowledge feature where agents retain useful snippets or sequences from earlier tasks, creating an internal playbook agents can query to avoid rediscovering strategies.

What This Means for Developers

Antigravity reframes the IDE as a governed environment for autonomous agents rather than a chat window with code actions. It anchors Gemini 3 Pro inside an IDE workflow, exposes distinct Editor and Manager views for supervising agents, and emphasizes task level visibility through Artifacts. The platform may accelerate complex automation and async collaboration while providing verifiable outputs that developers can review and refine.