In part 4 of a 5‑part series on building AI agents, Robyn Anderson (Jam) explains how to store agent definitions as markdown files in tools like Obsidian or Notion, making the agents platform‑agnostic and allowing memory to scale beyond the limits of any single AI runtime. She shows how to grant both Claude Code and Hermes access to a shared Drive folder (named “Sutton”) that contains these agent files. By selecting a specific agent—her Bozma marketing agent—she demonstrates calling it from the chat interface, generating a 7‑step client value sequence, and saving the result back to the folder. The episode emphasizes that this file‑based architecture lets agents continuously grow and learn, and teases the next episode on creating subagents and running multiple agents simultaneously.
Chapters
Welcome back to part 4 of 5 of how to build an AI agent. My name is Jam, and if we haven't met yet, I've trained hundreds of small business owners on how to use Claude and specifically build out their AI agents with no code. Okay, so we left off in part 3 where you were putting all of your information into your agent files. And if you're new here or just seeing this, agents are actually files on your computer or somewhere else. So I'm using Obsidian. You could use something like Notion. So it just depends where you want to keep your agent files. The reason we do this is it makes us platform agnostic. We are so in love with Claude right now, of course, but there's a bunch of other tools that you can use and different harnesses like Hermes and OpenClaw. And so we want to make sure that all of your content is available in the— in one spot. So you're not attached to a specific platform. Also, it makes the memory so much more— also, it makes the memory much bigger, so you can continue to add MD files for each agent rather than the platform itself. Okay, so now we got that out of the way. The agents are set up. Now it's time to pull the agent into your chat in Claude Code. So let me show you how to do that.
So whether I'm using Hermes up here or my Claude Code down here, I can give them both access to my Drive, which is my Obsidian folder, which I've named Sutton. So you can see here Sutton is called in, and all I did was just go to open folder and I put it on Sutton, which is my entire team. Or if I'm doing just marketing, I could just select that agent and it'll work just with that agent. So I called on my Bozma marketing agent and it created a 7-step client value sequence. And at the end, I can save it to my folder. So this is exactly why this is important, because otherwise this would just be a file in your system, right? Or if I'm doing this in Hermes, it would just be a conversation there. But using this agent file structure, it's constantly growing and learning from everything you do.
That's it for now. Tomorrow I'm going to show you how you create subagents and spin up multiple agents at one time. Continue to like, follow, and comment so that the algorithm can bring you back.
