coreyja.fm

Why I'm Starting a Podcast (and What I've Been Building)

March 6
22 mins

Episode Description

Welcome to the first episode of the coreyja.fm podcast! I'm Corey, I go by coreyja online, and I've been doing web dev for about 20 years. These days I'm focused on Rust and I recently took over leading the Battlesnake project.

This is a solo podcast about what I'm building, what's breaking, and what I'm learning. I've got a lot of side projects going and I want to do a better job sharing them with the world.

What I Cover

Project Tour — A quick rundown of everything I'm actively working on:

  • Mull — An AI agent orchestration and memory system. My "personal development environment" that remembers context across sessions and automates my development workflow.

  • Battlesnake — I took over hosting and maintenance of play.battlesnake.com at the start of 2026 and I'm rewriting the engine in Rust.

  • GAR — A tool to manage self-hosted GitHub Actions runners on a spare Mac, because my CI bills hit $100/month with all the AI-driven PRs.

  • Bake — Takes WordPress sites and turns them into static sites served from a single Rust binary. One server, multiple sites, all from memory.

  • Lavender Iguana — The business my wife and I run making WordPress sites for local businesses.

  • Small CLIs — Stamp (Fastmail integration), Porkbun (domain management), Quiver (skills manager), and more.

Deep Dive: The Mull Autopilot Pipeline — The main event. I walk through the full pipeline that takes a task from idea to pull request:

  1. Task quality check — A small model verifies the task has acceptance criteria and enough detail. If not, it launches an interactive session to interview you and flesh it out.

  2. Plan draft — An agent drafts an implementation plan from the task.

  3. Enrich — A dedicated step that searches the memory/learnings library for relevant past experience. This took memory utilization from ~3% to ~97%.

  4. Critique & revise loop — Different models critique and revise the plan (Gemini for critique, Opus for revision). Loops until approved or hits a max round count.

  5. Human review — The pipeline pauses here. You review the plan in markdown, edit it, chat about it, then approve.

  6. Implementation & review loop — Writes failing tests first, then implements and reviews in a loop until the review agent approves.

  7. Pull request — Opens a PR for final human review.

Over one weekend, this pipeline produced 42 merged PRs across 3 projects, with a 92% first-pass CI success rate. It's not magic — it's a productivity multiplier that lets me make progress on code while doing dishes or playing with my daughter.

Links

Next episode: Battlesnake — where it's at, where it's going, and the Rust rewrite. See you in two weeks!

See all episodes