Your first session, branches & models
Today you do your first real job with Claude. First, three words you'll use every day — each fits in one sentence.
Three words you'll use every day
Session
One conversation with Claude Code. Start a new session for each new task — it forgets, so fresh is good.
Branch
A safe copy of your project where Claude makes changes, so your main version is never broken.
Model
The "brain" you pick for the task. Smarter = slower and pricier. Choose per task (table below).
Open a working folder (Cowork)
In Cowork, Claude works inside a folder on your computer. That folder is its base: it can read what's inside and save what it produces there.
- In Cowork, start a new chat.
- Below the text box, look for "work in a project or folder".
- Create a new folder (for example
my-work) and select it. - Give it "always allow" so it can save changes without asking every time.
That's it: Claude now works inside that folder.
Your first real job
Try something concrete, not a chat question. For example:
Create a file called ideas.md and write 10 content ideas about my business. Here's what it's about: [describe your business].
You'll see Claude actually create the file in your folder. That's Cowork: it doesn't describe the file, it makes it.
The everyday loop
When you work on a repository with Claude Code (you connected it in the Connect GitHub lesson), the daily rhythm is four steps:
Start a session & pick your repo
New chat → choose the repository.
Work on a feature branch
Use or create a branch like feature/points. Claude also saves its work on its own branch, so you can review it before accepting anything.
Pick the model
See the table below. Switch any time with /model opus or /model sonnet — your chat is kept.
Paste a prompt & press Enter
Copy one from the library (the Prompts lesson), fill in the [brackets], send. Review the changes, then keep them.
Which model to pick?
Claude has several models, each with a different balance of speed and depth. The simple rule:
| Situation | What to pick |
|---|---|
| Quick, mechanical, or high-volume task | The fastest model |
| Work that needs reasoning, nuance, or important decisions | The most capable model |
You don't have to memorize names. Start with the default model; if a hard task falls short, move up to the most capable one. If a simple task feels slow, drop to the fastest.
And with names attached, here's the map (July 2026):
Sonnet 5
Fast and capable — most building and editing work.
Opus 4.8
Hard bugs, big refactors, planning, and reviews.
Haiku 4.5
Tiny edits and quick lookups.
opusplan
Opus plans, Sonnet builds.
Pro tip: use Opus to review and Sonnet to implement — reviewing is what needs the extra smarts.
The golden rule of sessions
Every working session starts "clean". The longer and messier a conversation gets, the less focused Claude's answers become. Two habits that make the difference:
- One big task = one session. Don't mix 5 different projects in the same chat.
- Give context up front, not in dribs and drabs. If Claude knows from the start who you are and what you want, everything that follows comes out better. (Next lesson we'll see how to give it that context permanently.)
Next lesson: skills — what turns Claude from a general assistant into a specialist in your field.