01What is an agent operating system and how it works

An agent operating system is a setup that runs your AI agents on a continuous loop. It is like a computer operating system, but for AI agents. It lets them work 24/7 without you needing to start them each time.

In this episode, Julian Goldie explains how an agent operating system differs from a simple goal. A goal is finite. It stops when the job is done.

An agent operating system loops. It keeps working. For example, Hermes Oracle runs every 24 hours.

It looks at headlines, pulls in tweets, and creates content ideas. Then it can draft social media content automatically.

You can set up this kind of system with tools like Hermes, Claude, or ChatGPT. They have scheduled tasks. These tasks keep running on a loop.

So you can have an agent that thinks of ideas, develops them, tests them, and improves them every day.

02Key features of an agent operating system

The agent operating system has many useful features. It can run continuous automation. It can also create custom agents for specific jobs like customer service.

You can describe what you want, and the system builds it.

  • Scheduled tasks: Run agents on a loop, like daily content creation.
  • Custom agents: Build agents for LinkedIn followups, SEO, or any task.
  • Mobile access: Use Tailscale or Cloudflare to access the system from your phone.
  • Token reduction: Use open-source skills to lower costs and avoid hitting limits.
  • Model switching: Switch between GPT-5.6 and GLM-5.2 to manage costs and performance.
  • Free API options: Use services like OpenRouter or a free AI coder with 90 providers.

These features make it easy to build powerful automation without being technical.

03How to set up your agent operating system

Setting up an agent operating system is simple. Julian shares that even non-technical people can do it. You just need to describe what you want and let the AI guide you.

  1. Install the agent OS from a zip file. You get a full guide and a tutorial video.
  2. Open your preferred AI tool, like Hermes, Claude, or ChatGPT.
  3. Ask it to create a scheduled task. For example, "Come up with an SEO content idea, create it, test it, and improve it every day."
  4. If you want to use your phone, set up a VPS with Tailscale or Cloudflare.
  5. If you want to avoid API fees, connect via OpenRouter or OSOA. You just log in.
  6. To reduce tokens, install open-source skills like Superpowers or use the token minimization playbook.

Julian says that creating an automation like a LinkedIn followup agent took only an hour or two with Fable 5. It was much easier than expected.

04How to use the agent operating system for real tasks

You can use an agent operating system for many tasks. Julian shows examples. One member built a LinkedIn followup agent.

Another created an AI chief of staff. These are just a few ideas.

  • SEO agent: Does keyword research, generates content, and deploys it across websites.
  • Video agent: Automates video creation and publishing.
  • Chief of staff: Starts each day with a briefing, keeps projects on track. You write a prompt and it runs daily.
  • News radar: Pulls latest news from Twitter or Google apps. You can change the category for any industry.

To use it, you just describe the task. For example, "Automate LinkedIn followups." Then the agent guides you step by step. You can also use a voice-activated version like Hermes Apollo.

It listens to your voice, builds things in real time, and can open apps on your computer.

05Results and comparisons from the episode

Julian shares results from building and using agent operating systems. One member, Sheena, built a LinkedIn followup agent. It took only an hour or two with Fable 5.

It runs on a loop automatically without needing constant attention.

Another member, Lee, created an AI chief of staff. It briefs them daily and keeps projects from falling through the cracks. This shows the system works for personal productivity.

Julian also compares Hermes and OpenClaw. He says they are basically the same, but Hermes is smoother and faster. He recommends sticking with one to avoid confusion.

When comparing GPT-5.6 and GLM-5.2, he notes GLM-5.2 is cheaper for coding tasks. You can use GPT-5.6 as the brain and delegate to GLM for specific jobs.

06Tips, limits, and mistakes to avoid

Avoid these mistakes when using an agent operating system. First, do not rely on goals that stop. Instead, use scheduled tasks that loop.

Second, do not use multiple models for everything. Use cheaper models for simple tasks to save tokens.

  • Token reduction: Use open-source skills like Superpowers, Caveman, or R2K. Change effort levels. Use older models like GPT-5.5 for simple tasks.
  • Delegation: Let GPT-5.6 orchestrate and delegate coding to GLM-5.2 to avoid burning tokens.
  • Mobile access: Use Tailscale or Cloudflare, but make sure you set up a VPS correctly.
  • Customization: When installing the zip file, ask the agent to change the news category to fit your industry.

Julian says you can reduce token usage by 90% or more with these tips. He also warns that frontier models like GPT-5.6 run out of tokens fast. So it is smart to have a backup like GLM-5.2.

07Who it is for and why it matters

An agent operating system is for anyone who wants to automate tasks without being technical. Julian says you do not need to be a coder. You just need to get started.

The AIPathy community shows many people building amazing things.

It matters because it saves time. You set up an automation once, and it runs forever. For example, you can automate SEO content, LinkedIn followups, or even a chief of staff.

This frees you to focus on other work.

Julian says that what took hours before now takes minutes. The system is available to community members. You get a full guide, video tutorials, and support.

If others can do it, you can too.