Anthropic's Claude Code brings AI coding to the terminal from $20/month

The article is a hands-on introduction to Claude Code, framed around the author's own confusion about how it differs from plain Claude. That distinction is the spine of the piece: Claude is for asking questions and consulting, while Claude Code is for handing over development work. The author emphasizes that AI code generation itself is nothing new, and that tools like Cursor, GitHub Copilot, and Codex also have agent features that inspect code, edit it, and run commands. The point, then, is not that Claude Code is unique, but that it represents a shift in how a developer interacts with AI — from conversation to delegation.

Much of the article then details the pieces that surround that delegation. CLAUDE.md holds project-wide rules; Skills capture procedures for specific tasks such as writing a Zenn article or doing code review; Permission Mode controls how much Claude Code may do without asking; Plan Mode lets it investigate and propose an approach before any changes are made; and MCP and connectors link it to outside services like Google Drive, Gmail, and GitHub. The author repeatedly notes that connectors usable in Claude and in Claude Code do not always match, and that operations depend on the connected service and authentication method.

On cost, the article describes a Pro and Max subscription path and an API pay-as-you-go path, and warns that Pro and Max are not fully unlimited — usage is shared with Claude, with 5-hour session and weekly limits. It also flags a practical trap: if an ANTHROPIC_API_KEY is set in the environment, Claude Code may authenticate with the API key and incur usage-based charges instead of drawing on the subscription. Whether the product fits a given developer may hinge on how those limits and billing paths line up with their actual workload.