What AntiGravity AgentOS Is and How It Works

AntiGravity AgentOS is a way to use the AntiGravity 2.0 app as an operating system for your work, not just a chat box. It groups your conversations into projects, gives agents memory files, lets them run on schedules, and uses skills to automate real tasks.

In this episode, Julian Goldie explains how Google's own team used the same pieces inside AntiGravity 2.0 to build a full operating system from scratch. They typed one prompt, and 93 agents worked together to make a working OS with a kernel, memory management, a file system, and even video and keyboard drivers. They played Doom on it. The older model, Gemini 3.1 Pro, could not do this at all.

The key idea is that you stop treating AntiGravity like a simple chat tool. Instead, you treat it like an OS that can run many agents at once, each with its own tasks and memory. That changes everything for how you work.

Main Features of AntiGravity 2.0

AntiGravity 2.0 is a standalone desktop app for Mac, Windows, and Linux. It no longer has a code editor or terminal inside it. You get a chat window with an agent, and that empty box can run up to 93 agents at once.

  • Projects group your conversations and can point at multiple folders. Each project has its own settings and permissions.
  • Memory reads rules files like agents.md, so every agent knows your business from the start.
  • Skills are folders with instructions. You write your process once, and agents can call it anytime.
  • Scheduled tasks let you type /schedule to run agents on a timer, like every morning at 9.
  • Hooks are another building block Google mentioned, though the episode does not explain them in detail.

Google also added a built-in skill called the AntiGravity Guide in June. It lets the agent read its own manual instead of guessing. The app updates every couple of weeks and tests new models in public.

How to Set Up AntiGravity AgentOS Step by Step

Setting up your own AgentOS is simple. Julian recommends a four-day plan, one hour a day. Here is the step-by-step process from the episode:

  1. Day one: Build your memory file. Write who you are, what you sell, who your customers are, how you talk, and what you never say. Half a page is enough.
  2. Day two: Create one project and point it at the folders that matter. For example, a content project can see your notes, brand, and website folders.
  3. Day three: Write one skill for the job you do most often. Keep the instructions clear and repeatable.
  4. Day four: Schedule that skill with /schedule. Set it to run daily or weekly.

That is your AgentOS. Start small. Do not try to build everything at once. The people who get stuck are the ones trying to do ten things on day one.

How to Use AntiGravity AgentOS in Your Work

Once you have your project set up, you can start using it for real tasks. Julian gives an example from his own work. He has a skill that writes SEO blog posts about AI automation for the boardroom. The skill follows his exact format and structure every time.

You can also use memory files to teach agents your business. Julian keeps his memory files in Obsidian because they stay on his machine and he owns them. He mentions that NotebookLM is great for turning notes into podcasts and videos, but for memory that plugs into agents, plain local files win.

Another example is the AntiGravity Guide. If you ask the agent how AntiGravity works, it reads its own manual instead of guessing. This shows how skills can make agents more reliable.

To schedule a task, you type /schedule and tell the agent to run once, tomorrow at 9, or every morning. You no longer need to manually invoke every agent. They work while you are away.

Results and Tests Shown in the Episode

The big test in the episode is Google's own experiment. They opened AntiGravity 2.0, typed one prompt, and asked agents to build an operating system from scratch. The result: 93 agents, a working OS, and they played Doom on it. Google's older model, Gemini 3.1 Pro, could not do it at all.

Google also said they did not use a secret internal version. They used the same building blocks that are in AntiGravity 2.0 right now: subagents, scheduled tasks, hooks, and projects.

However, not everyone liked the new version at first. Members in the community, like Alan and Gaz, said it was a major change. Jason came from Visual Studio Code and said he still needs an editor while he adjusts. Julian explains that the empty chat box feels worse, but it is actually more useful once you fill it with projects and skills.

In November 2025, AntiGravity was a code editor with AI bolted on. By May 2026, it became a standalone app that runs agent teams. In July, updates were shipping every couple of weeks.

Tips, Limits, and Mistakes to Avoid

One of the biggest mistakes is treating AntiGravity like a chat box. If you just type a question and copy the answer, you are not using it as an AgentOS. You need to set up projects, memory, and schedules to get real value.

Another mistake is trying to build everything at once. Connor, who is new to this, asked what to avoid. The answer: do not try to build the whole thing on day one. Start with one automation a week.

Also, if your system does not understand your business, the problem is not the tool. As member Dan said, you need to teach the system how your business works. Write a clear memory file.

Limits exist too. There is no code editor in AntiGravity 2.0. If you need one, you might have to use an external editor like Visual Studio Code while you adjust. The app is empty at first, but that is by design.

Keep your memory files local, like in Obsidian, so you own them. That gives you control over your data.

Who AntiGravity AgentOS Is For and Why It Matters

AntiGravity AgentOS is for anyone who wants to move from asking AI one question at a time to having a system that works on its own. This includes non-coders, because projects can point at folders full of Word docs and spreadsheets. You do not need to write code.

It matters because the future is about running agent teams, not typing prompts. Google is already moving in that direction. They are testing new models in public and adding features like scheduled tasks.

Julian also shares that if you want a pre-built AgentOS, you can get it inside the AI Profit Boardroom. It is a zip file that installs into AntiGravity with memory, skills, SEO setup, goals board, and Kanban board already built. There are coaching calls, a 30-day roadmap, a prompt library, and 2,800 business owners in the community.

To get started, you can go to aiprofitboardroom.com or join the free AI Success Lab with 75,000 members. The episode ends with a clear message: the same pieces Google used are on your desk tonight. Go set them up.