AI Engineering

MCP adds a hidden token and latency tax because models must read tool schemas and manage multi-step tool loops before they can answer.

Sandbox tasks vs code tasks, three types of automations, and how to build pipelines that span both environments from your phone.

A framework for understanding the tools you need to run Claude Code and Cowork from your iPhone, and how to debug when things break.

How to set up Tailscale, Termius, and Mosh for persistent Claude Code sessions from your phone, plus the Remote Control combo workflow that makes it practical.

AI instruction files work 70% of the time. Close the gap with hooks, automated validation, and team governance strategies.

Monorepos, legacy migrations, mixed framework versions. Here's how to configure AI coding assistants when your codebase doesn't follow one set of rules.

The mechanics behind phantom packages, version drift, and cross-module confusion, plus concrete prevention strategies for large codebases.

When AI takes architectural shortcuts your code still compiles, but breaks in production. Document the decisions AI can't infer from code alone.

Extra instructions don't help. They hurt. Learn to scope AI instruction files so your assistant only sees what matters for the current task.

Your AI assistant used the wrong framework? Here's how to write instruction files (CLAUDE.md, Cursor Rules) that actually prevent that.