Google DeepMind launched Anti-gravity, an agent-based integrated development platform that enables asynchronous coding through autonomous agents working locally to build and verify software projects. Featuring a familiar editor interface, an agent manager, and an integrated browser for real-time testing, Anti-gravity aims to revolutionize development by combining AI-driven automation with traditional coding workflows.
At the start of July, Google acquired the core team from the startup Windsurf, including its founders and key engineers, after a $3 billion acquisition deal with OpenAI fell through. Google DeepMind then integrated this team and, within just over four months, launched their first product: the Agentic development platform called Anti-gravity. This new platform represents Google’s entry into the agent-based integrated development environment (IDE) space, aiming to revolutionize how asynchronous coding and development are done by combining agent autonomy with traditional IDE features.
Anti-gravity is built around three main design pillars. The first is autonomy, emphasizing asynchronous work where agents operate locally on the user’s computer to handle tasks independently. The second pillar focuses on trust, verification, and self-improvement, where agents generate artifacts such as implementation plans and product requirement documents to guide and verify their work. The third pillar is the agent-first paradigm, encouraging developers to start with autonomous agents building the project and then transition to synchronous editing and manual code adjustments as needed.
The platform features a familiar editor interface similar to VS Code, allowing users to import existing settings and work comfortably. However, the core innovation lies in the agent manager, where users create and launch multiple agents that work asynchronously on different parts of a project. For example, the video demonstrated an agent building a Next.js app simulating gravity on Earth, Mars, the Moon, and anti-gravity, while another agent simultaneously checked the navigation bar. These agents produce detailed implementation plans and task lists, which users can modify on the fly, providing dynamic control over the development process.
Another unique aspect of Anti-gravity is its integrated browser with a special extension that agents use to test, verify, and interact with the application in real-time. This browser acts as both a simulator and a research tool, enabling agents to validate their work visually and gather external context when needed. The walkthrough feature offers a comprehensive summary of what the agents have done, including code diffs, screenshots, and instructions, making it easy for developers to understand and review the automated work.
Currently, Anti-gravity does not have a published pricing model, but Google plans to offer an extensive free tier with potential premium plans as the Gemini 3 model and other AI capabilities evolve. The platform shows promising advancements in agent-driven development, blending asynchronous AI assistance with traditional coding workflows. While some features resemble those of competitors like Cursor, Anti-gravity is a fresh product developed rapidly and internally refined over several months, signaling a significant step forward in AI-powered software development.