Claude Code Tutorial for Beginners: Your First Real Session [2026]

Your first Claude Code session decides whether you become someone who uses it daily or someone who tried it once. The difference is rarely the tool; it is whether anyone told you what is actually happening when the agent works, which requests set it up to succeed, and which five controls matter out of the dozens available. This tutorial is that first session, done properly: what Claude Code is, the install, a real task walked end to end, the agentic loop that runs underneath, and the smallest set of habits that make session two better than session one. No prior Claude experience assumed.

What Claude Code actually is

Claude Code is an AI agent that lives in your terminal and works inside your repository. You describe an outcome ("add input validation to the signup endpoint and cover it with tests"), and it reads the relevant files, makes a plan, edits code, runs your test suite, reads the failures, and iterates until the task is done or it needs your input. That distinction, an agent that acts and checks its own work, is what separates it from chat assistants that produce a code block for you to paste and debug yourself.

It sits alongside two other ways of working with Claude: the chat interface (thinking, drafting, one-off questions) and Cowork (agentic work over files in a workspace, without a terminal). The practical rule: if the work lives in a git repository and involves running things, Claude Code is the right surface.

Install and first launch

Installation is a single command from the official docs, and authentication happens on first run. Launch it from the root of a real project you know well, and resist the demo-repo instinct: the whole point of your first session is seeing the agent operate on code you can judge.

Your first real task, end to end

Pick a task that is real but bounded: a bug you already understand, a small refactor, a missing test. Then write the request with three ingredients a beginner usually omits:

1. The outcome, with success criteria. "Add validation to the signup endpoint: reject emails without an @, passwords under 10 characters. Return 422 with a field-level error body."
2. The verification. "Run the existing test suite; add tests for both rejection cases."
3. The boundaries. "Don't touch the user model or migrations."

Send it, and watch the loop: Claude explores the files it needs, states a short plan, edits, runs the tests, and fixes what fails. Your job during this is supervision. When it proposes a command or edit that needs approval, that is the permission model working: nothing state-changing happens without your consent until you choose to loosen it.

When the diff lands, review it the way you would review a junior colleague's PR, because that is the honest posture: fast, capable, needs review. Ask it to explain any change you don't follow ("why did you move the validator out of the handler?"), and you will learn the codebase's own logic back.

The five controls that matter on day one

Everything else can wait; these five cannot. Escape interrupts a run that is going sideways, no penalty, redirect and continue. Plan mode makes Claude propose a full plan for approval before touching anything, which is the right default for tasks over a few minutes. /context shows what is filling the context window (the working memory holding your conversation and files), which explains most "it forgot" moments. Checkpoints and rewind let you roll the session back to before a wrong turn instead of arguing with it. And CLAUDE.md, a file of project conventions Claude reads at session start, is one paragraph of setup that upgrades every future session: build commands, test commands, the rules of your repo.

What is actually happening: the agentic loop

Under every session runs one loop worth understanding, because it explains both the strengths and the failure modes. Claude reads your request plus available context, decides on an action (read a file, edit, run a command), observes the result, and repeats until it judges the task done. Two consequences follow. First, results depend heavily on what is in context, which is why CLAUDE.md and precise requests punch so far above their weight. Second, the agent's self-checks are only as good as the verification available to it: a repo with a test suite gives the loop something honest to push against, which is why "run the tests" belongs in almost every request.

Where beginners go next

After a few sessions, the growth path is predictable, and it is the ladder from the complete path guide: prompting patterns (Level 2), then CLAUDE.md and context management (Level 3), then plan-mode workflows and customization with skills and hooks (Levels 4 and 5). Developers heading toward certification will find the same skills examined by Anthropic's CCA-F; the practice tests show you the exam's angle on them.

Next steps

Do the free Level 1 today, then bring the habits back to your own repository. When you are ready for the full ladder with graded projects at every level, the course comes with Preporato Pro.

Sources:

• Claude Code documentation
• Claude Code quickstart
• Anthropic: Claude Code best practices