Claude Code, an AI agent, was tested on three real-world development tasks: a new feature in an internal admin dashboard built with Next.js, native macOS development for an Apple entitlements and signing process issue, and backend development in Go. The results showed that Claude Code handled some tasks impressively and stumbled on others, but its effectiveness was demonstrated by developers using it daily to tackle clear tasks in small codebases. The AI's performance varied across tasks, with some requiring significant human interaction and prompting to arrive at a solution. Lessons learned include the importance of providing specific test cases, enforcing Test-Driven-Development (TDD) principles, and controlling the work safely to maximize the value of using AI. Despite struggling with deeper reasoning and system complexity, Claude Code's performance improved with better prompts, more context, and internal tooling.