AI coding tools have become remarkably good at generating code, fixing errors, and explaining what happens inside a codebase. But Google has set its sights on something much larger: letting AI agents handle entire multi-step development tasks, not just individual lines of code.
Google’s answer is Antigravity, an agentic development platform designed to let AI agents take on substantial coding and knowledge-work assignments. These agents can use tools, work with files, search the web, delegate parts of a task to other agents, and keep going without needing constant instructions from a developer.
Antigravity first appeared in November 2025 as an AI-powered development environment. Since then, Google has expanded it into a broader platform that includes Antigravity 2.0, the Antigravity IDE, CLI, and SDK.
The core idea is straightforward: instead of relying on AI solely for individual code snippets, developers can assign agents larger objectives and then supervise the output. It is a shift from asking for a function to describing an outcome and letting the agent figure out the steps.
From Assistant to Agent
To understand the difference, compare Antigravity with a traditional coding assistant. A typical assistant might suggest code, explain an error, or generate a function on request. An agentic platform works at a higher level. You give it a broad goal, and it plans, uses tools, executes commands, and works through multiple steps to reach that goal.
Google describes Antigravity as a platform built for the agent-first era. Its agents can read and write files, run system commands, perform web searches, interact with Chrome, and create artifacts and implementation plans.
The platform now includes multiple surfaces, but the most significant change is the level of interaction. Instead of micromanaging every action, developers can describe the desired outcome and allow the agent to handle the execution.
Subagents, Background Work, and Schedules
Antigravity 2.0 introduces dynamic subagents. A main agent can create specialised agents for specific parts of a task, and these subagents can work in parallel. Workspace isolation keeps their work separate, which helps divide larger tasks into smaller pieces without cluttering the main agent’s context.
Asynchronous task management is also supported. Long-running operations can be moved to background processes so they don’t block active work. Subagents can run in the background while their progress streams back to the main agent.
Scheduled Tasks add another layer of automation. Users can set recurring schedules that automatically trigger agents to perform predefined jobs. Google offers examples like daily pull-request digests, hourly checks on live deployments, and monthly reports on system architecture changes. Once configured, agents operate without manual prompting.
Artifacts: Seeing What Agents Actually Do
One of Antigravity’s most important features is Artifacts. These are outputs agents create to show their work and progress. They can be implementation plans, rich documents, diagrams, images, browser recordings, or other evidence of what the agent has done. This addresses a critical challenge: when AI handles more work, how do you know what it actually did? Artifacts give users something to inspect rather than relying on a final answer alone. Google has designed Antigravity around this idea, allowing users to review the work and provide direct feedback.
Powered by Gemini
Antigravity is tightly integrated with Google’s Gemini models. The current platform highlights Gemini 3.7 Flash, which launched in August 2026 as the workhorse model for coding and agentic tasks. Google says the model shows notable gains over its predecessor, Gemini 3.6 Flash, on coding benchmarks, and it comes with introductory pricing at roughly half the cost per token.
However, Antigravity is not simply a model. The model supplies 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 matters: Antigravity is a platform for deploying and managing agents, not just another chatbot powered by Gemini.
A Changing Role for Developers
The bigger shift is from AI that helps write code to AI agents that take on larger pieces of work. Traditional development requires developers to decide what needs to be done, write or modify code, run tests, investigate problems, and repeat. Agentic development changes that balance. A developer can increasingly describe an outcome and let an agent handle more of the execution, leaving the developer to direct the work, review results, and make decisions that require human judgment.
This does not mean developers disappear. It means the developer’s role shifts from manually carrying out every step to orchestrating, supervising, and reviewing AI agents. Google is positioning Antigravity squarely around that evolution.
Control and Safety
Autonomy comes with risk, and Google has built in permission and security controls. By default, Antigravity requires interactive approval before agents run terminal commands. Project-level permissions can restrict access to files and other resources. These safeguards are essential because an agent that can modify files, execute commands, or interact with external tools needs clear boundaries. The result is closer to human-supervised autonomy than fully unsupervised AI.
Pricing and Availability
Google currently offers Antigravity free of charge under its Individual plan. The company frames this as public-preview pricing, and usage quotas have already been adjusted several times since launch. Paid Google AI plans offer higher usage limits and priority access. Antigravity 2.0 runs on Windows, macOS, and Linux. The free availability is significant because it lowers the barrier for developers who want to experiment with agentic development without immediately committing financially.
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 handle substantial work. With Antigravity 2.0, developers get a command centre for managing agents, plus a terminal interface, a programmable SDK, and a dedicated IDE. Agents can work in parallel, operate asynchronously, delegate tasks, use tools, and run on schedules. The challenge ahead is no longer simply whether AI can generate code, but whether developers can direct, supervise, and trust agents that execute increasingly complex work. That is what makes Antigravity worth watching.
