Claude Code

Claude Code is the most leverage-heavy tool I’ve added to my stack since the GPU. It’s also the easiest to use badly. The difference is structural: skills as first-class capabilities, hooks that codify your team’s invariants (“always run tests before commit,” “never push without typecheck”), subagents for parallel work, and a plan-then-execute discipline that keeps the agent from compounding its own mistakes.

Posts under this topic cover real workflows — the brainstorming → spec → plan → execute arc, hook configurations I use across every repo, how to write a skill that the model actually invokes, when to use subagents versus when to keep work in the main thread — and the failure modes that show up when teams treat Claude Code like an autocomplete with a chat box.

If you want a starting point, the AI Native Development topic page covers the broader philosophy; this one is the practitioner detail.