Making the web faster and more user-friendly
Google Antigravity: The New IDE That Embraces Agentic AI
There’s a new IDE from Google that embraces agentic AI and leverages Gemini 3: Google Antigravity.
With Gemini 3's advanced reasoning, tooling, and autonomous programming capabilities, Google Antigravity transforms AI assistance from a tool in a developer's toolkit to an active partner. While the core of Google Antigravity is a familiar AI IDE experience, its agents have been elevated to a dedicated interface and gained direct access to the editor, terminal, and browser. Now agents can autonomously plan and execute complex, end-to-end software tasks on your behalf, while validating their own code.
On Friday, my workmate Brian and I put it to work on some complex task in the development of the project we're working on, with the Gemini 3 model, and we were quite impressed by it.
What I love about Antigravity is that it handles planning by producing two text artifacts, just like a Tech Lead would: a Walkthrough and Tasks. You can provide feedback on these — whether it made a mistake or if the prompt wasn't specific enough — by selecting text and commenting on it, similar to how you would in Google Docs or Notion. You can also give feedback on the entire approach, much like in a Pull Request. I almost always write "LGTM" (Looks Good To Me) there. Then it starts working. Actually, lately, it starts even before your feedback to get a head start, confident in its plan, and can correct course later if needed.
I tried working without Antigravity, using only OpenSpec with Gemini 3 via the gemini-cli, but it felt more cumbersome and frustrating. The CLI isn't asynchronous like Antigravity, so you lose a lot of time waiting for the AI to think. Antigravity breaks everything down into smaller tasks that are easier and faster for the AI to handle, so it effectively starts working sooner and finishes faster.
Then there's the Agent Manager. If you feel like being an AI-powered superhero coder, you can ask it to do several things in parallel, even on different repos (or different clones of the same repo if you need to work on two different branches).
I wanted to know more on how to use it best, so this weekend I took some time to watch a couple of videos that I highly recommend.
Learn the basics of Google Antigravity #
The first one is "Learn the basics of Google Antigravity" from the Antigravity team.
This video provides an official introduction and walkthrough of Anti-gravity's core components:
- Three Main Surfaces: The tool combines an Agent Manager (to create and manage agents), an Editor (a full IDE experience), and a Browser (incorporating an agent directly inside Chrome for autonomous testing).
- Agent Assisted Development: The LLM is set up to automatically make decisions and implement easy tasks. For complex issues, it can pause and prompt the user for clarification.
- Artifacts for Transparency: The agent generates markdown files like Implementation Plans and Walkthroughs to document its progress and changes.
- Parallel Task Execution: Developers can kick off a task for the agent and start another complex task themselves.
- Autonomous Testing: The agent can launch the integrated browser to verify its own code changes.
Google Antigravity: The New Best AI Editor? #
If you want a more in-depth view, check out "Google Antigravity: The New Best AI Editor?" from "AI with Brandon".
This review highlights:
- Agent Manager Inbox: The main interface is designed like an email inbox for managing multiple projects.
- Free Gemini 3 Pro Access: Includes free access to Gemini 3 Pro within the editor.
- Complex Task Handling: The reviewer demonstrates the agent successfully implementing a full-stack feature involving multiple external services (trigger.dev, OpenAI TTS, blob store, database, and UI updates).
It looks like it's gonna change the way we develop, by turning us into... orchestrators of agents?