AI coding tools have become good at generating code, fixing errors, and explaining what is happening inside a codebase. But Google wants its AI to take on a larger role in the development process.

That is where Antigravity comes in.

Antigravity is Google’s agentic development platform, designed to allow AI agents to handle multi-step coding and knowledge-work tasks. The agents can use tools, work with files, search the web, delegate parts of a task to other agents, and continue working without needing constant instructions from a developer.

Google introduced Antigravity in November 2025 as an AI-powered development environment. Since then, the company has expanded it into a broader platform that includes Antigravity 2.0, the Antigravity IDE, CLI, and SDK.

The idea is simple: instead of using AI only to help with individual pieces of code, developers can give agents larger tasks and then supervise the work they produce.

This article explains what Antigravity is, how it works, and why it matters as software development becomes increasingly agentic.

What is Google Antigravity?

Google Antigravity is an agentic development platform that allows users to work with AI agents on coding and other knowledge-work tasks.

The easiest way to understand the difference is to compare Antigravity with a traditional coding assistant.

A coding assistant might suggest code, explain an error, or generate a function when asked. An agentic platform works at a higher level. Users can give an agent a broader objective and allow it to plan, use tools, execute commands, and work through multiple steps.

Antigravity is Google’s entry into an already crowded field. Tools such as GitHub Copilot, Cursor, Anthropic’s Claude Code, and OpenAI’s Codex have all moved beyond autocomplete and chat toward agentic modes that plan and execute multi-step coding work. What distinguishes Antigravity is that it is built as an agent-first platform from the ground up, spanning a desktop command centre, terminal, IDE, and SDK, rather than adding agent capabilities on top of an assistant product.

Google describes Antigravity as a platform built for the agent-first era. Its agents can perform tasks such as reading and writing files, executing system commands, conducting web searches, interacting with Chrome, and creating artifacts and implementation plans.

The platform now has several different surfaces:

• Antigravity 2.0: A standalone desktop command centre for launching, monitoring, and orchestrating agents.
• Antigravity CLI: A terminal-based interface for working with Antigravity agents.
• Antigravity SDK: A Python framework for building custom agentic applications on top of Google’s Antigravity runtime.
• Antigravity IDE: A full agentic development environment designed for working directly with code.

How does Google Antigravity work?

The biggest difference with Antigravity is the level at which users interact with the AI. Instead of asking an agent to write one function, a user can give it a broader task and allow it to determine the steps required to complete it.

Antigravity agents can use tools to execute commands, read and write files, search the web, interact with Chrome, and work with external services through skills and MCP servers. MCP, the Model Context Protocol, is an open standard for connecting AI models to external tools and data sources, and its presence means Antigravity agents can draw on a widening ecosystem of integrations rather than only the tools Google ships itself.

This changes the relationship between the developer and the AI. The developer does not necessarily have to specify every individual step. They can describe the desired outcome and let the agent work through the task.

Antigravity 2.0 supports dynamic subagents, which allow a main agent to create specialized agents for particular parts of a task.

These subagents can work in parallel, with workspace isolation helping keep their work separate. Google says this can allow larger tasks to be divided into smaller pieces while preventing multiple workstreams from unnecessarily filling the main agent’s context.

In practical terms, instead of one agent doing everything sequentially, it can delegate focused pieces of work to other agents.

Antigravity also supports asynchronous task management. Long-running operations can be moved into background processes so they do not block the agent’s active work. Subagents can also run as background tasks while their progress is streamed back to the main agent.

Antigravity 2.0 also includes Scheduled Tasks. Users can set recurring schedules that automatically invoke agents to perform predefined tasks. Google gives examples including daily pull-request digests, hourly checks on live deployments, and monthly reports on system architecture changes.

This takes Antigravity beyond an AI tool that users manually prompt every time. Once a schedule has been configured, the agent can be triggered automatically according to that schedule.

What are Antigravity Artifacts?

One of the more important parts of Antigravity is Artifacts.

Artifacts are outputs that agents create to communicate their work and progress to the user. They can include implementation plans, rich documents, diagrams, images, browser recordings, and other forms of evidence about what the agent has done. This matters because handing more work to an AI agent creates another problem: how do users know what it actually did?

Artifacts give users something to inspect rather than forcing them to rely solely on a final answer. Google has built Antigravity around this idea of communicating progress and results through artifacts, with users able to review the work and provide feedback directly.

What models power Antigravity?

Antigravity is closely integrated with Google’s Gemini models, the same model family that underpins the company’s wider AI products.

Google’s current Antigravity platform highlights Gemini 3.7 Flash, introduced in August 2026 as the platform’s workhorse model for coding and agentic tasks. Google says the model shows notable gains over its predecessor, Gemini 3.6 Flash, on coding benchmarks, along with introductory pricing at roughly half the cost per token.

Antigravity is not simply a model, though. The model provides the underlying intelligence, while the Antigravity agent harness provides the environment, tools, permissions, and other capabilities that allow an agent to carry out multi-step work.

That distinction is important. Antigravity is a platform for deploying and managing agents, rather than simply another chatbot powered by Gemini.

Why does Antigravity matter?

The bigger change is the move from AI that helps developers write code to AI agents that can take on larger pieces of work.

Traditional software development requires developers to decide what needs to be done, write or modify code, run tests, investigate problems, and repeat the process. Agentic development changes that balance.

A developer can increasingly describe an outcome and allow an agent to handle more of the execution. The developer then becomes responsible for directing the work, reviewing the results, and making the decisions that require human judgment.

That does not mean developers disappear. It means the developer’s role can shift from manually carrying out every step toward orchestrating, supervising, and reviewing AI agents.

This shift is not happening at Google alone — it is the direction the entire developer-tools industry is moving in, with every major vendor racing to add agent capabilities. Google is clearly positioning Antigravity around that shift. The company describes the platform as a way to work in the agent-first era and has expanded it beyond coding toward broader knowledge work.

Can Antigravity work without constant supervision?

Antigravity is designed to support autonomous and asynchronous work, but that does not mean users should blindly hand over control.

Google has built permission and security controls into the platform. For example, Antigravity’s default settings require interactive approval before agents run terminal commands, while access to files and other resources can be restricted through project-level permissions. This is important because greater agent autonomy also creates greater risk.

An agent that can modify files, execute commands, or interact with external tools needs appropriate boundaries. Antigravity therefore combines autonomy with controls that let users determine what agents are allowed to access and do. The result is closer to human-supervised autonomy than completely unsupervised AI.

Is Google Antigravity free?

Google currently lists Antigravity as available at no charge for developers under its “Individual” plan. This is framed as public-preview pricing, however, and usage quotas have been adjusted more than once since launch. The company also offers paid Google AI plans with higher usage limits and priority access.

Antigravity 2.0 is available for Windows, macOS, and Linux. The free availability is significant because it lowers the barrier for developers who want to experiment with agentic development, at least for now, without immediately committing to a paid platform. That contrasts with the subscription tiers common across rival agentic coding tools, though the public-preview framing means the terms could continue to change.

Bottom line

Google Antigravity matters because it captures one of the biggest changes happening in software development: the move from AI as an assistant to AI as an agent that can take on larger pieces of work.

Antigravity 2.0 gives users a command centre for managing those agents. The wider ecosystem adds a terminal interface, a programmable SDK, and a dedicated agentic IDE. Agents can work in parallel, operate asynchronously, delegate tasks, use tools, and run on schedules.

The challenge now is no longer simply whether AI can generate code. It is whether developers can direct, supervise, and trust agents that can execute increasingly complex work. That is what makes Antigravity worth watching.