Claude Code is a terminal-native agentic coding tool built on a minimal-scaffolding philosophy: give the model filesystem access, a bash shell, and a loop, then get out of the way. Built with TypeScript, React, Ink, and Bun, it implements a three-phase agent loop (gather context, take action, verify results) with deliberately few tools. The team deletes scaffolding with every model upgrade, letting model improvements translate directly into product improvements. Key context engineering levers include CLAUDE.md (persistent project memory), skills (reusable instruction sets), and hooks (automated pre/post-action rules). Research data complicates the productivity narrative: a METR study found experienced developers were 19% slower with AI tools while believing they were 20% faster, and Faros AI found PR volume nearly doubled while review times increased 91%. The gains are real for individual developers on well-scoped work but collapse at the organizational level without review automation infrastructure. The competitive moat is the model itself, not the product layer, making context engineering the primary differentiator for users.