Claude Code — The Planner-Executor

Claude Code is Anthropic's agentic CLI, and it occupies a specific seat in my workflow: the planner-executor. Where other agents shine at speed or scale, Claude Code shines at depth — holding a whole codebase in mind, planning multi-step changes, and executing them with surgical precision across dozens of files.

What makes it different

• Plans before it touches code. For any non-trivial task it produces an explicit plan I can correct — the cheapest possible place to fix a mistake.
• Whole-codebase edits. Search, understand, then edit coherently — not file-by-file blind patches.
• Tool discipline. It shows its work: commands run, tests executed, diffs staged.
• Taste. In practice, its refactors read like a careful senior engineer wrote them, which minimizes my review surface.

This portfolio is a Claude Code build: a full VitePress theme clone adapted to two locales — planned once, reviewed by me, shipped in this shape.

How I brief it

The better the plan I can critique, the better the result. So my briefs for Claude Code state the goal, the constraints, the definition of done — and explicitly invite a plan before execution. Then I do what the human is for: I read the plan, attack the weak points, and only then let it run. The pattern is described in The Agentic Method.

Claude Code also plays the integrator role after Codex fan-outs: reconciling diffs, resolving the overlaps, and running the full suite before anything reaches a branch I care about.