A former Meta and Microsoft principal engineer named Kun Chen now merges 40 pull requests per day without manually reviewing a single line of code. He built three free, open-source tools to make this work: Lavish, an HTML-based visual planning editor; Treehouse, a system for running 20 to 30 parallel coding agents simultaneously; and No Mistakes, an automated AI code review tool that catches errors before they reach production.
The workflow follows three stages: plan, code, validate. The planning stage is where most engineers fail with AI. Kun uses HTML artifacts instead of markdown because they render structured, interactive specs that agents parse more reliably. He walks through the exact process of turning a rough idea into an AI-ready spec, which is the part of the conversation most worth watching closely.
The full episode includes live demos of each tool, a breakdown of what Kun actually checks before merging AI-written code at minute 45, and a direct answer to how someone builds competency in agentic engineering from scratch. All three tools are available now on GitHub under kunchenguid.
[WATCH ON YOUTUBE →]