Antigravity

in English: Google Antigravity

advancedspreading

Google's developer program in which AI agents independently plan, write, and test software, while humans assign tasks and review their work.

In plain terms

Antigravity is a programming environment designed by Google so that AI agents perform most of the work. It does not just suggest the next line of code; you can assign it an entire task, such as "create a booking page."

It is like when on a construction site you no longer lay the bricks yourself but act as a site manager. You assign work to the crews, they work, and you review the plans and completed parts.

Unlike traditional coding assistants, multiple agents can work in parallel and operate beyond the editor. They execute commands and test what they build in a browser.

They provide clear 'evidence' of their work: task lists, implementation plans, and screenshots. This way, you do not need to read every line to see where they are heading.

How it works

Describe in everyday language what you want: a new feature, a bug fix, or a complete small application.
The agent uses a background language model to create a plan, break the task into steps, and show it to you for approval or comments.
It then works independently: it creates and edits files, runs commands in the terminal, and then opens and tests the result in a browser.
It creates so-called artifacts about the work (plan, task list, screenshot, browser recording) to which you can add comments, like a shared document.
On a separate dashboard, you can launch and monitor multiple agents simultaneously, while in the traditional editor view, you can also intervene in the code yourself.

An everyday example

A small business developer assigns three tasks in Antigravity in the morning: fixing a faulty form, creating a new price list page, and updating tests. While the developer is in a meeting, agents work in parallel. Upon returning, they review the plans and screenshots, accepting only what is correct.

Who is affected

developerYour work is shifting increasingly from typing code lines to assigning tasks, supervising agents, and verifying results. → Code

entrepreneur, executiveA prototype or internal tool can be developed faster with a smaller team, but oversight and responsibility remain with you.

jobseeker, new graduateSome entry-level programming tasks are taken over by agents, making the ability to review plans and spot errors more valuable. → Work & HR

parent, student, teacherWhen learning programming, it is easy to build a working application with it, but without understanding, your knowledge has nothing to rest on. → Education & Family

A common misconception

MythWith Antigravity, anyone can build finished, reliable software without programming knowledge.

In realityAgents deliver quick, impressive results but can make mistakes, misinterpret tasks, and leave security vulnerabilities in the code. Using their output in production without expert review is risky.

Limits and risks

An agent can run commands and delete files on your machine, so consider how much autonomy you grant it and ensure you have backups.
Hidden instructions found on the internet or in files can mislead the agent (prompt injection), which can lead to data leaks.
Your code and instructions are sent to cloud-based models, so for corporate or confidential projects, check the data processing terms first.
Usage has quantitative limits, and pricing and the range of available models may change, so do not base long-term plans on current conditions.

Similar, but not the same

Gemini – Gemini is the language model itself, while Antigravity is the work environment where the model operates as an agent. In Antigravity, models from other companies can also be selected alongside Gemini.
Code completion – Code completion suggests lines while you type, whereas Antigravity agents carry out entire tasks from planning to testing.
Gemini CLI – Gemini CLI is a coding agent running in the command line, while Antigravity provides a full graphical development environment with an interface for managing multiple agents.

Related concepts

Broader

AI-agent, coding agent, Integrated development environment (IDE)

Narrower

Artifact, Agent management interface

On the edge

Cursor, Claude Code, Windsurf, Vibe coding, Gemini

A brief history

On November 18, 2025, Google introduced Antigravity alongside the Gemini 3 model as a free public preview for Windows, macOS, and Linux. The program is built on the open-source foundation of the Microsoft Visual Studio Code editor.

Where things stand

A tool known and widely tested among developers, competing with Cursor, Claude Code, and other coding agents. Few people outside the developer world use it yet.

Status: 2026 November

In the news

Appears in 2 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Google's AI agent can propose new drug molecules within hours, DeepMind saysNov 3, 2026 · Health
Agent platform race: Cursor and OpenAI coordinators, Atlassian, AWS, Cloudflare, Moonshot, AntigravitySep 21, 2026 · Code

Also known as

Antigravity · Google Antigravity · Antigravity IDE · Google Antigravity IDE · Google's agentic development environment · agent-first IDE · Google coding agent

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.