I'm starting from scratch here and documenting as I go. I've set up my initial framework (DOE) on Antigravity and built a simple reviewer agent that reviews all code the parent agent builds. Nothing fancy yet, the goal is just to get something running end-to-end. When I say "I built", I mean I said to Claude Code, "Build a reviewer agent that does xyz". A.I. is getting mental! So far, this is my basic tech stack: - Google Antigravity (IDE) - Claude Code (Agent/Engineer) - Wispr Flow (Voice prompts) - Framework (Directive, Orchestration, Execution, aka D.O.E.) What's working: - Agent runs - The reviewer agent reviews the code and rates it out of 100. What's broken / in progress: - When the rating is below a certain threshold, I want the parent agent to autonomously improve the code. Essentially, taking me out of the picture, which allows me to focus on the next biggest needle-moving activity. I'm intentionally keeping this simple. The goal is not a perfect system yet; it's to learn through building, reviewing failures, and then iterating. BUILDING IS THE FASTEST AND 'FUNNEST' WAY TO LEARN! Just as AI requires feedback loops to get smarter. Humans require feedback loops to learn. I decided to still post this even though the build's incomplete. If you're a beginner, this is what "starting" looks like. Next steps: - Update the Directive layer prompt with instructions to always run the reviewer agent after the parent agent has built code. - Update the reviewer agent to instruct the parent agent to improve code autonomously if the rating is below a set threshold (e.g., 70)

This could be helpful for anybody who likes to multitask while coding. I built a notification sound ("ping") to play whenever Claude Code finished a task or completed a response. I did this to make sure I always came back to the chat as it responded. I built this by simply prompting Claude something like: "Play a ping sound notification whenever you complete a task or finish your response". It's literally that easy to build nowadays! I noticed that when creating a new chat instance (/new), the notification sound stopped working. To fix this, all I did was informally ask Claude Code: "Why isn't the ping sound playing when you complete a task?" It then immediately diagnosed the problem, fixed the coding, fixed the directive prompts, and re-tested all in about 45 seconds. The lesson: It's 2026, when your tools have AI built in, debugging is a conversation, not a search engine marathon. Welcome to the future lads. 😎