Your first session

Time for the actual first conversation. We're going to do this slowly and on purpose so you feel the rhythm of a Claude Code session. Nothing impressive — just enough to make the mechanics click.

Set the stage

Make a fresh folder somewhere — somewhere you don't mind being a little messy. The folder will be the project Claude can see. Then open a terminal in that folder and run:

mkdir ~/Projects/claude-first-session
cd ~/Projects/claude-first-session
claude

After a moment, the terminal will switch into Claude Code's interface — a clean prompt at the bottom, a thin status bar above it, and a welcome line. The cursor is waiting for you to type.

Your first message

A good first message is something small and concrete enough that you can watch Claude do every step. Try:

Hi. Please create a file called hello.txtwith the text "Hello from my first Claude Code session" inside it. Then list the files in this folder so I can see it worked.

Press Enter and watch what happens.

You'll see something like:

1. Claude prints a short reply: "Sure — I'll create the file then list the folder."
2. A box pops up asking permission to write to hello.txt. Approve it (we'll cover permissions properly next lesson).
3. Claude writes the file. You see a small confirmation.
4. Another permission prompt — this time to run ls. Approve.
5. The output of ls appears, showing hello.txt.
6. Claude wraps up: "Done. The file is created. Anything else?"

That's a complete agent loop. Read, act, observe, report.

The shape of a good first task

Good early tasks share a few qualities:

• Small. Two or three steps, not twenty.
• Concrete."Make a file called X with text Y" — no ambiguity.
• Verifiable. You can check whether it worked without expert knowledge.
• Recoverable.If Claude does the wrong thing, it's easy to undo (delete a file, etc.).

Save the "build me a whole web app" tasks for later. Build trust on the small stuff first.

Try a slightly bigger one

Now ask:

Create a folder called notes, and inside it create three Markdown files: monday.md, tuesday.md, and wednesday.md. Each one should have a heading with the day and a single paragraph below it saying "Notes for this day go here."

Watch Claude break this into steps: maybe one tool call to create the folder, then three tool calls to create the files, then a summary. Approve as it asks. When it's done, open the folder in your file explorer and confirm everything is where it should be.

Common first-session mistakes

Asking a vague question

"Make me something cool." Claude will try, but you'll get something generic. The clearer your goal, the better the output. Even "make me a static landing page for a coffee shop" is enormously better than "make a website."

Not reading the diff

When Claude edits files, it shows you the changes. Look at them. Especially in your first sessions. This is the habit that catches problems early.

Treating it like Google

Claude Code can answer questions, but if you're just looking something up, you're leaving most of its value on the table. Claude shines when there's real context — your files, your codebase, your real problem.

Ending a session

When you're done, press Ctrl + D or type /exit. Claude saves a transcript and shuts down cleanly. The next time you run claude in the same folder, you can pick up where you left off — or start fresh.

What just happened

In about ten minutes, you used an AI agent to do real work in real folders on a real machine. You watched it think, ask permission, act, report. You started to develop a feel for what kind of request gets what kind of result.

That feel — the sense of what to ask for and how — is the single most important thing to build over the next module. Every session is practice.