01What an Agent Operating System Is and How It Works

An agent operating system is a single dashboard where you can run all your AI agents in one place. Instead of switching between 14 different tabs or terminals, you open one system and control everything from there. This makes your AI work faster and more organized.

The system connects agents like Claude, Gemini, N2, and others so they can share memory and outputs. When one agent finishes a task, its result feeds automatically into the next agent. You don't need to copy and paste between models anymore.

In this episode, Julian Goldie shows how he built an agent operating system out of Claude without writing code, and he uses it daily to build games, SEO pipelines, presentations, and more.

02Key Features of an Agent Operating System

  • Group chat for agents – You can have multiple AI agents talk to each other in a single group chat.
  • Persistent memory – Every conversation from every agent is auto-logged in an Obsidian vault, so agents never forget what you discussed.
  • Workspace gallery – Every image, video, audio file, or HTML page you create is auto-saved to a workspace tab, not lost in a downloads folder.
  • Voice-activated assistant – Hermes Jarvis can be set to wall mode with a wake word, and it will give you daily briefings.
  • Cron job scheduling – You can schedule tasks to run while you are away and manage them in one place.
  • Plug-and-play model updates – When a new AI model releases, you simply add it to the system. If a model gets banned, you plug in a better one without rebuilding.

03How to Get Started with Your Own Agent Operating System

  1. Get the system files. Julian offers his setup inside the Airpuffer Ballroom classroom. Go to airpufferballroom.com and find the Agent OS section.
  2. Download the zip file that contains the agent operating system build.
  3. Watch the video tutorial that comes with the zip file to learn how to install and configure it.
  4. Add your first AI agent by putting in its API key. The transcript mentions free APIs for Claude code, N2, Gemini, and Grok build work with the system.
  5. Set up your workspace folder where all created files will auto-save.
  6. Configure persistent memory by linking the Obsidian vault so every conversation gets logged automatically.
  7. Test the system by asking an agent to complete a simple task, like generating a landing page or a daily briefing.

04How to Use the Agent Operating System: Examples from the Episode

In the episode, Julian shows several ways to use the agent operating system. First, he uses the group chat feature where multiple AI agents talk together to solve problems. He also shows Paperclip, a full organization of AI agents that can build projects together.

Julian built seven different projects this way.

He demonstrates a voice-activated Hermes Jarvis set to wall mode. He asks it for a daily briefing, and while he works on other tasks, the agent researches and compiles the latest news into a presentation. The final files drop into his workspace automatically.

Another use is the SEO deployment pipeline. He types in a keyword and a case study, and with one click, the system deploys content to all his websites. He also built an RPG open world game and a Skyrim-style open world game using the same system.

05Results and Comparisons Shown in the Episode

Julian compares the old way of using AI to the agent operating system. The old way means juggling 14 different tabs, copying and pasting between models, and losing files in random downloads folders. With the agent operating system, everything is in one tab, outputs flow automatically between agents, and every file auto-saves to a workspace you can revisit later.

He also shares results from Fusion, a new system that uses five models and one judge to fuse answers together. On benchmarks, Fusion scored 69% on a Draco test and beat Fable 5 by 65%. When Fable 5 got banned, the agent operating system let him plug in Fusion instead without any downtime.

06Tips, Limits, and Mistakes to Avoid

  • Stay anti-fragile – Do not rely on any single AI model. If one gets banned or stops working, your system should let you plug in another model immediately.
  • Avoid cold starts – Do not open a terminal and wait for an agent to load every time. Use the dashboard to click on an agent and see its history right away.
  • Do not lose your creations – If you use chatbots directly, files end up in the downloads folder and you cannot find them later. Use a workspace that auto-saves everything.
  • Update the system daily – Julian updates his system every day. To keep getting better results, you should add new models and features as they release.
  • Non-coders can succeed – Julian saw people like Ethan, age 61 and not a coder, build their own agent operating system. You do not need to code to get results.

07Who Should Use an Agent Operating System and Why It Matters

This system is for anyone who uses multiple AI tools and wants to stop juggling tabs. It is especially useful for business owners, content creators, and developers who need to automate workflows like SEO deployment, video creation, or daily research briefings. In this episode, Julian explains that non-technical people like a 61-year-old named Ethan built the system and got great results.

Using an agent operating system matters because it makes AI truly useful. Instead of AI making you faster but your to-do list still not done, this system helps you finish tasks. You can schedule work to run while you sleep, and every new model update makes the whole system better, not harder to manage.

As Julian says, you own the system, not the models.