Google Antigravity is probably the most misunderstood code editor in the AI space right now. It looks like a chat interface, so most people treat it like one. That’s a mistake. After three months of using it seriously, across real client work and personal projects, I’ve found a handful of features that have genuinely changed how I work — and I want to walk through each one.
1. The Agent Manager
When you first open Antigravity, there’s a button front and centre that grabs your attention. Ignore it. The thing you should open first is the Agent Manager, tucked in the top-right corner.
The Agent Manager is what separates Antigravity from every other AI coding tool. It lets you run multiple conversations across multiple projects — simultaneously. Not switching between tabs. Actual parallel execution.
In practice, this means I can:
- Spin up a brand new website in one workspace
- Ask a legacy codebase to explain itself and propose a feature plan in another
- Write a blog article for my wife’s book site in a third
All at the same time. It’s a mission-control view of your entire development workload. Each workspace is just a folder on your hard drive, but the Agent Manager turns that collection of folders into something you can genuinely orchestrate.
This is the feature I’d recommend understanding before anything else.
2. Planning Mode
Most AI coding tools have one mode: execute. You type something, the model writes code. That works well for small, contained changes. But at the start of a new project, it’s a recipe for going in the wrong direction fast — burning tokens and generating code you’ll throw away.
Antigravity’s Planning Mode changes this. Instead of immediately writing files, the model produces an implementation plan first. That plan is saved as a markdown artifact inside the tool, and here’s where it gets interesting: you can comment on it.
Not in a “send another message” way. Literally, in-line comments — like you’re reviewing a colleague’s Google Doc.
I used it recently to plan a Three.js space shooter. The model asked whether I wanted a top-down or third-person perspective. It flagged the technology choice (vanilla vs React Three Fiber). I replied directly in the comments: top-down, vanilla, prioritise performance, add juice (screen shake, pop animations). Those comments fed back into the plan, and only then did it start writing code.
The efficiency gain is real. A planning pass writes a markdown file. A code pass writes dozens of files. Starting with the plan means you’re steering before you’re committed, using a fraction of the tokens.
3. The Playground
Sometimes you don’t want a workspace. You don’t want a folder on your hard drive. You just want to think out loud with an AI and maybe produce a file or two.
That’s what the Playground is for.
It functions like a lightweight version of ChatGPT, except — critically — it has access to your file system. So when I needed marketing copy for a photography website, I didn’t just get a wall of text back. I told it to write the copy to a markdown file, and it did. That’s a genuinely different capability from a browser-based chatbot.
The Playground creates a temporary folder implicitly. Delete the conversation, and the files go with it. If you decide the work is worth keeping, there’s a single button to promote it into a permanent workspace folder.
4. The Changes Panel
Inside the Agent Manager, there’s a Changes Panel (top-right corner, “toggle changes panel”) that shows every artifact the agent has created or modified.
The artifact system is what gives Antigravity its editing model. Instead of generating a new version of a document each time you ask for a change, it edits in place. You see the current state of the file, comment on specific parts, and the model modifies exactly what you pointed at — nothing else.
When you use Planning Mode, the Changes Panel becomes a project management view. You get:
- An implementation plan — the architectural overview
- A task list — the breakdown of work
- A walkthrough — a record of every change made
That’s the kind of documentation a senior software engineer produces before a major feature. Antigravity produces it automatically, as a byproduct of working the way I’ve described.
5. Autocomplete
This one is more familiar — it works similarly to GitHub Copilot. As you write code in the editor, Antigravity suggests completions in italics. Tab to accept.
Where it gets interesting is the Snooze button, buried in the bottom-right corner of the editor under Antigravity settings.
Autocomplete is genuinely useful when you’re in flow and you want the model to handle boilerplate. But sometimes you want to think with your own brain, without a persistent grey suggestion nudging you toward something you didn’t ask for. Snooze turns it off temporarily.
I use Snooze more than I expected. It’s a small thing, but having the control makes the tool feel like a tool rather than a system that’s always trying to finish your sentences.
6. AI-Generated Commit Messages
The last feature is simple, and I use it almost every day.
At the end of a working session — especially a long one with a lot of small changes across many files — it can be genuinely difficult to remember what you’ve done. Antigravity has a Generate Commit Message button that reviews your changed files, considers the context of your recent conversations, and writes a meaningful commit message.
It sounds trivial. In practice, it’s one of those features that saves you from writing “update stuff” at 11pm.
The Bigger Picture
Taken individually, none of these features is magic. The Agent Manager is a project switcher. Planning Mode is a structured prompt. The Playground is a scratch pad. But together, they create a workflow that feels fundamentally different from how I used to build software.
The thing that’s shifted most isn’t speed — it’s confidence. Having a plan before committing to code, being able to comment on that plan like a document, running multiple workstreams without losing context — it changes how you approach a project from the start.
If you’ve been treating Antigravity like a chatbot, spend a week working with the Agent Manager open and Planning Mode on by default. I’d be surprised if it doesn’t change your view of what this tool actually is.
Found a feature I missed? Drop it in the comments — I’m still learning this thing.