Claude Code

What It Does

Claude Code runs in a terminal (or as an SDK/IDE integration, desktop app, or browser) and works like a pair programmer with direct access to a project's files and shell. It can read a codebase, plan and make multi-file edits, run tests and linters, fix failures it finds, and execute git operations — all while asking for permission before risky actions.

Key Features

• Agentic workflow: Plans a task, edits multiple files, runs the test suite, and iterates on failures without needing every step spelled out. "Dynamic workflows" can orchestrate many parallel subagents for larger tasks.
• Shell and tool access: Can run arbitrary terminal commands (with permission prompts) to build, test, and inspect a project.
• Current models: Defaults to Claude Opus 5 on Max/Team Premium/Enterprise plans and the API, and Claude Sonnet 5 on Pro/Team Standard seats. Claude Fable 5 is available (not default) for the hardest, longest-running work via /model fable. Effort levels (low/medium/high/xhigh/max) let you trade cost against reasoning depth.
• Extensibility: Supports custom subagents, skills, hooks, and MCP (Model Context Protocol) connections to external tools and data sources.
• IDE and CI integration: Available as a CLI, VS Code/JetBrains extension, desktop app, browser (claude.ai/code), and headless mode for CI pipelines and automation.

Who It's For

Software engineers who want an AI agent that can autonomously execute coding tasks — refactors, bug fixes, test-writing, migrations — rather than just suggest inline completions.

Practical Tips

• Give it a CLAUDE.md file with project conventions to reduce back-and-forth.
• Use plan mode for larger changes before letting it edit files directly.

Limitations

Requires a Claude subscription (Pro from $17-20/mo, Max from $100/mo) or API credits — there's no fully free standalone tier. Best results come from clear task scoping; very open-ended prompts can lead to overbroad changes.