Lesson Content

Code mode is not just for developers.

The name "Code" implies software development, and Code mode is genuinely powerful for developers. But the name undersells it. Code mode is more accurately described as Claude's agentic mode – a version of Claude that can take actions on your computer, not just respond to questions about it.

Understanding what "agentic" means is the key to understanding this mode.

What agentic means.

In Chat mode, Claude is a respondent. You ask, Claude answers. The output is text you read and decide what to do with.

In Code mode, Claude is an agent. It can take actions: read a file on your hard drive, write or edit that file, run a terminal command, search the web, spawn sub-agents to work on parallel tasks, and chain these actions together across multiple steps to complete a complex goal. You describe what you want done; Claude figures out how to do it and does it.

This is a fundamentally different relationship with the tool. Chat mode is conversation. Code mode is delegation.

What Code mode can do that Chat mode cannot.

• Read and write files: With your permission, Code can read documents, spreadsheets, code files, and any other file on your computer directly – without you copying and pasting their contents. It can also create new files and edit existing ones.
• Run shell commands: Code can execute terminal commands – running scripts, installing packages, moving files, querying databases, checking system status.
• Use external tools: Through a system called Model Context Protocol (MCP), Code can connect to external services, databases, APIs, and applications configured on your computer.
• Spawn sub-agents: For large tasks, Code can create multiple parallel agents, each working on a piece of the problem independently, then merge the results.
• Multi-step task execution: Code can plan a sequence of actions and execute them one by one, checking its own work and adjusting as it goes.

CLAUDE.md – the project memory file.

When you work in Code mode on a specific project or directory, you can create a file called CLAUDE.md in that location. Claude reads this file at the start of every Code session in that directory. It is where you write standing instructions: how this project is structured, conventions to follow, commands that are commonly needed, things Claude should always or never do in this context.

CLAUDE.md is how you teach Code mode about your specific world – so you stop re-explaining the same context every session. We cover CLAUDE.md in full detail in the Persistent Memory course.

Who gets real value from Code mode.

Software developers: The most direct use case. Code can review files in a codebase, write and edit code, run tests, manage git operations, and work through development tasks step by step.

Technical writers and documentation authors: Code can read source files and generate documentation from them directly, maintaining accuracy between code and docs.

Data analysts and researchers: Code can read data files, run analysis scripts, and produce summaries – without requiring the analyst to manually export and paste data.

Power users with complex file workflows: Anyone who works with large numbers of files – organizing, processing, converting, summarizing – can delegate that work to Code mode rather than doing it manually.

Professionals using Claude Code as a daily assistant: With CLAUDE.md set up well, Code mode remembers your preferences, your projects, and your working style across every session. For heavy Claude users, this persistent context is transformative.

The permission model.

Code mode operates with a permission system. When Claude wants to take an action – read a file, run a command, edit something – it tells you what it intends to do and asks for approval. You can approve individual actions, approve all actions of a type for the session, or deny. This permission model is what makes Code mode safe to use even for powerful operations: Claude does not act without your knowledge.