What happens when you take the human out of the agent loop. Building an autonomous issue-to-PR pipeline in Rust with MCP, fail-open design, and filesystem IPC.
What I learned setting up a large-scale iOS codebase so that AI coding agents can work in it effectively, from hierarchical documentation to executable skills.
How we built a skill that lets AI coding agents verify their own iOS changes on a real simulator, with video recording and accessibility auditing.
Why wrapping your iOS build commands in a Makefile is the highest-leverage change you can make for AI coding agents.
How I rewrote Claude-XcodePreviews in Swift and accidentally built a declaration-level dependency resolver.
How building a native macOS GitHub notification client exposed surprising gaps in GitHub's public API and stress-tested Lists in a real app.
How to set up an iOS project for git worktrees, including DerivedData isolation, SPM caching, simulator conflicts, and project file merge strategies.
What I learned building claude-deep-review, a code review skill that decomposes review into 15 specialized agents running in parallel.
What I learned building Claudio, a tool that runs multiple AI coding agents simultaneously using git worktrees
What it's actually like to build a real tool on Apple's on-device FoundationModels framework — the 4,096 token ceiling, blind token budgeting, and everything in between
What I learned building a CLI toolset that captures SwiftUI previews programmatically, giving AI coding assistants a visual feedback loop.