Google has built its Antigravity coding agent into Google AI Studio, turning the browser-based tool from a place to prototype with Gemini into one that can build and run full-stack web apps from a text prompt.
The upgrade was announced on 18 March 2026 by Ammaar Reshi and Kat Kampf of the AI Studio team. Google calls it a “full-stack vibe coding experience”: the user describes an app in plain language, and the agent writes the code, adds the libraries it needs and, with the user’s approval, connects a backend through Firebase.
What Antigravity can now do inside Google AI Studio
According to Google’s announcement, the agent in AI Studio’s Build mode can:
- detect when an app needs a database or sign-in and, after the user approves a Firebase integration, set up Cloud Firestore for storage and Firebase Authentication for “Sign in with Google”;
- build real-time multiplayer apps and shared workspaces;
- install third-party web libraries, such as Framer Motion for animation or shadcn for interface components;
- store a user’s own API keys for outside services, such as payment processors or Google Maps, in a new Secrets Manager in the Settings tab;
- keep a project’s state across sessions and devices; and
- build apps in React, Angular and, newly, Next.js.
Google said the agent now keeps a deeper understanding of the whole project structure and chat history, which it says allows more precise multi-step edits. It also said the new experience “has already been used internally to build hundreds of thousands of apps over the last few months”. That figure comes from Google and has not been independently verified. Google said it was working on Workspace integrations for Drive and Sheets and on a one-click route to move an app from AI Studio into the full Antigravity environment.
Antigravity launched in November 2025 as a separate agentic IDE
Antigravity itself is not new. Google introduced it on 20 November 2025 as a desktop “agentic development platform” for macOS, Windows and Linux, released in public preview at no cost for individuals. At launch it offered Gemini 3 Pro with what Google called generous rate limits, plus support for Anthropic’s Claude Sonnet 4.5 and OpenAI’s open-weight GPT-OSS.
Antigravity has two main views. The Editor View is a conventional AI-assisted code editor with tab completion and inline commands. The Manager Surface is where a developer can start, coordinate and watch several agents working at the same time across different workspaces. Google says the agents can plan and carry out a task across the editor, the terminal and a browser: for example, writing a feature, launching the app in the terminal and then testing it in the browser.
Google has also used the agent outside software. Google DeepMind has shown Antigravity helping to find a new antimicrobial peptide candidate, which still needs laboratory testing.
Antigravity’s Artifacts aim at the review problem in AI coding
The most useful idea in Antigravity is not speed but review. Instead of asking developers to scroll through raw tool calls and terminal logs, the agents produce what Google calls Artifacts: task lists, implementation plans, screenshots and browser recordings. A developer can comment on an Artifact, much as on a shared document, and the agent takes the feedback into account without stopping its work.
That matters because the bottleneck in AI-assisted programming has shifted from writing code to checking it. When an agent can change dozens of files in minutes, a person still has to decide whether the change is correct, secure and maintainable. Other teams have reached the same conclusion: Anthropic has described how its Claude Code team reviews code as agent output surges. Artifacts do not remove the need for that judgement, but they give reviewers a summary they can test against, such as a recording of the feature working in a browser.
The AI Studio upgrade lowers the barrier further, which brings benefits and risks. People without programming experience can now build a working app with sign-in and a database. But apps that store user data and API keys need sound security rules and careful handling of secrets, and a non-specialist may not know what to check. Google’s design, which asks for approval before connecting Firebase and keeps keys in a dedicated store rather than in code, addresses part of that risk. Whether the default security settings hold up in real use is something independent testing, not Google’s launch post, will show. Larger companies are already experimenting: Epic Games has said its new games launcher, now in preview, was “vibe coded”.
Google’s offer competes with Cursor, GitHub Copilot, Replit and others, many of which also let users pick between models from several labs. Our comparison of AI coding tools sets out how they differ.
Update, 30 October 2026: Google is winding down Jules, its asynchronous coding agent, and says it is bringing what it learned into Antigravity. Here are the dates and how to export your Jules data.
Update, 2 November 2026: this article has been revised. It previously dated the AI Studio upgrade to 19 March 2026 (Google announced it on 18 March) and listed Claude Sonnet 4.6, Claude Opus 4.6 and Gemini 3 Pro as the models supported, which does not match Google’s launch post. Investment commentary, unsourced job-market forecasts and pricing claims have been removed.
Related Reading
- Google Is Shutting Jules; Sessions End 11 November, Export Opens Now
- AI Agents Are Learning to Spend Money: The Payments Infrastructure Race
Sources
- Introducing the new full-stack vibe coding experience in Google AI Studio — Google, 18 March 2026
- Build with Google Antigravity, our new agentic development platform — Google Developers Blog, 20 November 2025
- Getting started with Google Antigravity — Google Codelabs
Get TSN’s biggest tech stories, explained.
News and insights across AI, tech, crypto and cybersecurity. Every source linked. Unsubscribe anytime.
Get the newsletterFollow @tsnmediahub on X for breaking AI news as it happens.
Follow on X
