How I use Claude Code without becoming its passenger

Short summary

An engineer shares how they use Claude Code as an autonomous coding agent while staying in control — structuring problems precisely, reading every diff, interrogating the agent's choices, and owning outcomes. They also reflect on the atrophy of hands-on coding skills and course-correct by hand-writing code when sprints allow. The core insight: leverage demands skill, and accountability is what separates using a tool from being its passenger.

• •Treat AI agents like a fast, confident junior: structure problems, read diffs, interrogate choices, own outcomes
• •Unattended agents produce 'almost right' code that compiles but fails silently in production
• •Hand-write code when possible to prevent skill atrophy from review-only workflows

Generated with AI, which can make mistakes.