5 Things I Bet You Didn't Know About Antigravity
By AI LABS
Summary
Topics Covered
- Nano Banana generates UI styles
- Agents Manager orchestrates multi-agent workflows
- Isolated browser enables grounded research
- Browser simulates real user testing
- Planning mode creates feedback loops
Full Transcript
Along with the release of Gemini 3, Google also dropped anti-gravity. Initially, I thought it was just another VS Code fork, but we saw that Google shipped a lot of other interesting features as well. The new Nano Banana Pro is one of them, and it's not only an anti-gravity, but Google Stitch as well. It's free for now across all tiers as it's a preview version, but the rate limits exhaust pretty fast. It does have potential, though. So, let's dive into five anti-gravity features you need to
know. With the nano banana integration, you can ask the model to generate the UI style you want your website to follow. I asked Anti-Gravity to generate three vintage themed website designs for a photography portfolio, each with its own unique style. I used the second one, and it created the website with this style. It was a pretty minimal website with great fonts, and it used blank placeholder images. The fonts it used matched the editorial aesthetic really
know. With the nano banana integration, you can ask the model to generate the UI style you want your website to follow. I asked Anti-Gravity to generate three vintage themed website designs for a photography portfolio, each with its own unique style. I used the second one, and it created the website with this style. It was a pretty minimal website with great fonts, and it used blank placeholder images. The fonts it used matched the editorial aesthetic really
well, giving it an overall clean look. The design wasn't exactly the same, but it still looked good enough. The crazy part is you can even ask Nano Banana to generate images for your website, which I did to fill those placeholder images. The agents manager is another interesting feature. You can use the agents view and the editor view side by side with multiple agents working together. The idea behind it is to give manager-like control over tasks. You can have multiple agents working
simultaneously within a project and choose between them. Each agent works in the same project unlike others where they operate in isolated work trees. There are workspaces which are essentially separate projects. So you can have multiple projects open at the same time with agents running in each following their own set of tasks. The whole manager runs in the background and there's an inbox where you'll get notifications whenever an agent finishes its work. You can check and update their
progress. With agents manager, you can assign different tasks to different agents and have them work on various areas of your website. You can follow along with each agent and watch as it makes changes in real time. When you install anti-gravity for the first time, you'll be prompted to install the anti-gravity browser extension because it also has a browser integration. It's not builtin and it runs on a separate Chrome profile so your agents don't interfere with your actual browser
progress. With agents manager, you can assign different tasks to different agents and have them work on various areas of your website. You can follow along with each agent and watch as it makes changes in real time. When you install anti-gravity for the first time, you'll be prompted to install the anti-gravity browser extension because it also has a browser integration. It's not builtin and it runs on a separate Chrome profile so your agents don't interfere with your actual browser
activity. This browser also enables the model to automatically launch and do proper well-grounded research for you. I asked it to look up popular themes for photography websites and after asking if I wanted to allow cookies, it continued working in the browser. It also provided a full recording after completion and it serves as an artifact for the model to understand the process. This brings us to the most interesting feature in anti-gravity. It uses the browser to test the web apps simulating
activity. This browser also enables the model to automatically launch and do proper well-grounded research for you. I asked it to look up popular themes for photography websites and after asking if I wanted to allow cookies, it continued working in the browser. It also provided a full recording after completion and it serves as an artifact for the model to understand the process. This brings us to the most interesting feature in anti-gravity. It uses the browser to test the web apps simulating
interactions with scrolls, button clicks, and other interactions to detect issues. It takes screenshots and recordings while testing. When I asked it to check my website, it navigated through all the pages and took numerous screenshots to identify where the application was lacking. The testing and fixing took considerable time. Although it didn't fix the padding issues on the sidebar and navbar, by the end, the final website had improved a lot. When you give any prompt to the model to
create something in planning mode, it first generates an extensive implementation plan and then waits for you to review it. Once you've refined or reviewed the plan, you can click proceed so that the model can start the implementation. With this comments feature, you can comment on the implementation plan or any other artifacts it creates, and it will incorporate those changes directly into the implementation. Not only that, you can also add comments midway while it's
executing, making for a more natural feedback loop than the usual chat other agents have. With a [snorts] few upgrades and optimization, I could honestly see this as becoming my new daily coding agent. That brings us to the end of this video. If you'd like to support the channel and help us keep making videos like this, you can do so by using the super thanks button below.
Loading video analysis...