Antigravity

Antigravity is a program from Google that people use to develop software. It belongs to the so-called development environments: programs in which one writes program code, runs it, and searches for errors. What is special about Antigravity is that a large part of the work is done by artificial intelligence. The user describes in ordinary language what should be created, for example a website with a login form. The AI then writes the code itself, runs it, and checks in the browser whether it really works. Google unveiled Antigravity in November 2025, together with the AI model Gemini 3.

From autocomplete to independent work

Previous AI assistants for programmers mainly suggested the next lines of code. The human remained constantly at the wheel and had to accept or reject every suggestion. Antigravity shifts this division of roles. The human sets a goal and then checks the result. The many small steps in between are handled by the software.

For the industry, this is interesting for two reasons. On the one hand, the work of developers changes: less typing, more checking and planning. On the other hand, such a tool is a battleground between the major providers. Google is competing here directly with Microsoft's GitHub Copilot, with Cursor, and with Anthropic’s Claude Code. Whoever provides the environment in which people program daily binds users closely to its own AI models.

Antigravity is free to use at launch, though with limits on the number of requests. This pricing policy is typical for an early market phase. Providers initially accept losses in order to shape habits.

Agents, manager view, and artifacts

Technically, Antigravity is built on Visual Studio Code, a very widespread free development environment. On top of this, Google deploys so-called agents: AI programs that break a task down into sub-steps and work through them one after another. An agent can create files, execute commands in the terminal, and remotely control a browser. This allows it to click through the page it just built itself and see whether the button responds.

New is the manager view. There, several agents run in parallel on different tasks, and the user observes them like a project manager overseeing their team. Instead of a long wall of text, the agents deliver so-called artifacts: a task list, a plan, screenshots of the test. These pieces of evidence make it faster to judge whether the work holds up.

Antigravity is not tied to a single AI model. Besides Gemini 3, Claude from Anthropic and open models can also be selected. A common misconception is that Antigravity itself is an AI. It is rather the workshop in which various AI models are put to work as tools.

What this means for users and the market

For pupils and students, Antigravity is above all a way to quickly build small projects. A website, a simple game, an evaluation script for spreadsheets. The program is available for download for Windows, macOS, and Linux. Anyone who uses it should nevertheless be able to read the generated code, since agents make mistakes and occasionally invent functions that do not actually exist.

In business news, Antigravity usually appears as evidence of a larger trend. Experts speak of agentic programming, that is, of AI that not only suggests but acts. Analysts estimate that at large software companies, a considerable share of new code is now generated by machine. Such figures are hard to verify, but they shape expectations on the stock market.

Security questions in particular are being critically discussed. An agent with access to a terminal and a browser can accidentally delete data or fall for manipulated websites. Shortly after launch, security researchers reported the first vulnerabilities of this kind. Companies therefore often introduce such tools only with tight permissions and clear approval rules.