@zav_developer's reel — transcript & breakdown

The video starts with a personal introduction and a bold claim about having a superior AI development setup, immediately engaging the audience.

“I'm a 23-year-old AI engineer in big tech, and this is my AI development setup. P.S. My setup is better than yours.”

The speaker introduces the core components of his AI development stack, setting the stage for the detailed demonstration.

“I have Pi as my agent harness. I have ChatGPT as the model provider using the $20 subscription. I have Herdr, which helps me manage multiple agents at the same time. And I do all of this in a terminal called WezTerm.”

He delves into the customizability of Pi and demonstrates a free transcription feature, highlighting unique benefits.

“I really like Pi because of its customizability. You can change certain aspects that you just absolutely cannot change with other providers. When I load up Pi, you can see all of my context. You can see all of the skills that I have, and I'll go through those in a second. You can see the extensions that I have. This is what makes Pi unique, and you can see the themes. So one of the first things that I can do with my Pi setup is I can use transcribe natively in Pi for free. This is completely free. Hello, Pi. What are you up to today? As you can see, that totally worked. I send the message. It will respond. There you go.”

The speaker showcases his terminal-centric workflow, including Neovim and TUI for calorie tracking and Spotify, emphasizing efficiency and customization.

“So everything happens in the terminal. I've got Neovim. I track all my calories and protein for the day using a TUI. I even do Spotify with a TUI. I like a terminal-centric workflow where I can navigate around with keybinds. I can, you know, do terminal multiplexing. I can switch around to different workspaces, etc. So I really like TUI.”

He explains AXI, an agent framework, and its benefits for streamlining CLI interactions, showcasing a practical example with GitHub.

“AXI is an agent framework that was developed by Kun Chen, who is a principal engineer at Meta and a few other companies. AXI is a framework that is supposed to replace MCP. It prioritizes making TOON-formatted CLIs that wrap capabilities. So instead of using the GitHub CLI, you create an AXI wrapper around it and use that instead. Instead of using the Exa MCP, you create an AXI around it and use that instead. So that's what these skills are. Use Exa AXI to do some searching. Search for Kun Chen's GitHub. There you go. So this is the guy who invented AXI. And it got returned results in the TOON formatting, which is much more token-efficient than regular JSON.”

The speaker details additional skills and tools within his setup, like Unslop and Chrome DevTools AXI, highlighting their specific functionalities.

“I have a few other skills, like how to programmatically work with Herdr. Unslop, which makes the agent respond in a way that isn't horrible. Chrome DevTools AXI allows it to play with a browser. And then AXI is just the skill for knowing what AXI is and how to make CLIs using it.”

He demonstrates a 'Toggle Skills' extension, explaining its importance for managing agent context efficiently and giving credit to its creator.

“As you can see, my Pi is very pretty, and it does these little thinking things just like Claude does. And I've sort of customized the interface to my liking. So one of the extensions I have is called Toggle Skills. And what this does is it allows me to toggle which skills are agent-invocable or manual-only. That being said, this extension is not created by me. It was created by a guy named Dillon Mulroy, whose GitHub can be seen right here. This is important because you don't want all skills to be invocable by an agent. You don't want to clog the agent's context up. You only want it to be able to see the skills that you want it to use autonomously.”