Google’s Anti-gravity is a phenomenal AI coding tool, but its true power is only unlocked when you use it correctly. I’ve discovered a workflow that will have you building AI applications at an unprecedented speed. If you follow along with this guide, you’ll not only master Anti-gravity but also build a complete application, even if you’ve never written a line of code.
Let’s dive in.
Getting Started with Anti-gravity
First, you need to get the tool. Anti-gravity is a brand new, completely free AI coding tool. After a quick installation, you’ll be greeted by a screen that might look familiar if you’ve used other AI code editors.
However, we won’t be spending our time on that main screen. The real magic of Anti-gravity lies within the Agent Manager. This is where the core of our workflow will take place. Opening the Agent Manager reveals a powerful interface for managing workspaces, agents, and the playground—features that are incredibly unique to this tool.
Building Our First Application
To demonstrate this master workflow, we will build a stock investing and news application. This app will allow us to track companies, view their stock prices, see the latest news, and get AI-powered analysis.
To begin, open a new workspace and create a new folder for our project. This action will spawn a new agent dedicated to that workspace. We’ll provide this agent with a specific prompt.
A couple of things are crucial when working with Anti-gravity, especially with Gemini 3. You need to provide UI inspiration. A great way to do this is by using a tool like V0 to generate a UI based on a description. You can then take a screenshot of the generated interface and paste it directly into your agent as visual inspiration. Gemini 3 excels at building UIs, but only when given a visual guide.
We also need to select the right mode. Anti-gravity offers “Planning” and “Fast” modes. Interestingly, “Planning” mode still writes code, so it’s more of a “slow and thoughtful” mode. To ensure it only generates a plan, we’ll explicitly tell it not to write any code.
Here is the prompt we will use:
I want to build an investment web app using Next.js and Tailwind CSS. I need to be able to enter a stock ticker, and the app should display a price chart and the latest news for that company. Use the attached screenshot as visual inspiration for the app’s design. Do not write any code; just build the plan for me.
With the prompt ready, we’ll send it to the agent and let it get to work.
The Power of Artifacts
One of the most awesome features of Anti-gravity is how it creates artifacts as it works. As the agent builds the implementation plan, you can click to open it while the agent continues its tasks. It generates an entire document outlining the plan.
Unlike other AI builders that simply list plans in the console, Anti-gravity treats everything it builds as a distinct artifact. This means you have reviewable documents for every step of the process. It is crucial to read these artifacts as the agent works, as it makes understanding the subsequent steps much easier.
Interactive Testing and Annotation
Once the plan is executed, the agent doesn’t just stop. It begins testing the application as it builds. You’ll see an integrated browser where the agent opens the app, tests its functionality, and takes screenshots. While other AI app builders require you to test everything manually, Anti-gravity integrates full end-to-end testing.
But it gets even better. You can interact with this process. For instance, if you want to add a new section, you can highlight an area on the screenshot and leave a comment.
Let’s say we want a crypto section. We can simply highlight the sidebar and add a comment: “Add a crypto section here.” The agent will then add that task to its queue based on your annotation. This is an incredibly powerful feature that no other tool currently offers.
After a short while, the first version of our app is ready. It looks clean and functional for a V1. The agent even has full control over a browser instance, clicking through features and testing everything for you. It’s the closest we’ve come to having a true AI employee.
Scaling with Multiple Agents
That initial build is just a fraction of what makes Anti-gravity unique. Now, let’s leverage the Agent Manager to do more. We can spin up multiple agents to handle different tasks simultaneously.
-
Product Manager Agent: Click the plus sign next to our project and create a new agent. We’ll assign it the role of a product manager with the following prompt: “You are a product manager agent. Take a look at our app and build me a roadmap of more features we can build.”
-
Marketing Agent: Let’s create another agent for marketing. The prompt will be: “You are a marketing agent. Write me a blog post and five tweets announcing our new app and all its features.”
-
Feature Development Agent: We can also spin up another agent to work on a new feature. Let’s tell it: “Allow us to add the amount of shares we own for each stock so we can track our portfolio’s value and daily performance.”
Now we have multiple agents working in parallel on different aspects of our project.
Experimenting with Playgrounds
As you work, new ideas will inevitably pop up. Instead of cluttering your workspace with new projects, you can use Playgrounds. Playgrounds are for experiments and new ideas.
For example, I’m a big gamer and just had an idea for a 3D first-person shooter. Instead of creating a new project, I’ll open a playground and give it this prompt:
I want to build a 3D first-person shooter game using 3.js. Make it cool and stylistic.
The agent will start working on this experiment in the playground without interrupting our main project. If the idea proves promising, we can easily move it to its own workspace later.
The Inbox: Your Command Center
With multiple agents running, how do you keep track of everything? The Inbox. This is where your agents communicate with you, providing status updates and asking for approvals. It’s like an email inbox for your AI team.
Here, we can see that our portfolio tracking agent has finished its work. Our product manager agent has created a roadmap artifact. And our marketing agent has drafted a blog post and tweets. We can review each of these and approve the work.
The Mindset Shift: From Builder to Manager
This workflow requires a fundamental mindset shift. You are no longer just a person building an app; you are a manager of a one-person business. You have AI employees for marketing, product management, and development. You have an R&D division experimenting with new ideas in the playground. And you have an inbox to manage all communications.
It’s a shift from hands-on coding to high-level management. This is the future of building with AI.
Finalizing the App
Our agent has finished building the portfolio tracking feature. The app now has a “Holdings” section where you can enter the number of shares you own for each stock. After saving, the app displays your new portfolio size and the total value, showing how much it has changed that day.
And just as we were testing that, the agent has already started working on the crypto dashboard we requested via annotation. It automatically picked up the next task. After a quick approval, the app now has a crypto option in the sidebar with real-time crypto prices.
This is how you achieve incredible speed. You manage agents, annotate their work in real-time, and spin up new experiments as ideas come to you. You’ve transitioned from a hands-on guide to a full-on manager.
Conclusion
While there are still some rough edges Google needs to smooth out with Anti-gravity and Gemini 3, particularly around instruction following, the workflow is undeniably powerful. By using multiple agents in workspaces, running experiments in playgrounds, and managing everything through your inbox, you can achieve a level of productivity that was previously unimaginable.
Keep the Agent Manager open, and throughout your day, manage your team of AI agents. You will get so much more done.