Claude Code in six steps

Step 1 · Get the app

Five minutes, one download

Download the official Claude desktop app and install it like any other program. The page below has both versions; pick macOS or Windows.

You may have heard Claude Code described as a tool for programmers in a black terminal window. That version exists. You will not use it. The desktop app gives you the same engine with a normal interface.

Step 2 · Sign in and subscribe

Account, plan, and the two-minute model tour

Create an account with your personal email and take the Pro plan. It covers everything in this guide.

The model names, decoded. You will see a menu of models. They are engine sizes, nothing more. Click each one once, then leave the default selected and move on.

One more dial: thinking effort. Next to the model name you will find the effort setting (it might say "High"). It controls how long Claude reasons before it answers.

Beginner rule for both dials: leave the defaults alone. Raise them only when an answer to a hard question felt shallow, and drop back after.

Step 3 · Open Claude Code, make your test folder

Tell Claude who you are, once

Inside the desktop app, two ways of working sit side by side, switched by the two small icons at the top of the window: a chat bubble, and code brackets </>. Click the brackets. That is Claude Code: the mode that works inside a folder on your computer.

The two icons at the top: chat bubble and the marked </> brackets. Click the brackets.

Create a normal folder on your computer called test. Then, just under the message box, click the folder chip and choose that folder as your workspace. From now on, everything you and Claude produce together has a home you can see in Finder or Explorer.

The folder chip (marked) is where you choose your test folder. Below: the Manual approval mode, the microphone, and on the right the model and effort settings.

Now paste this as your first message, filled in with your own details. Claude will save it as its standing instructions for this folder, so you never introduce yourself twice. Prefer talking? Press the microphone in the bar under the message box and dictate instead of typing; the words land the same way.

Save these as your standing instructions for this project.

Who I am: [name, role, industry, years of experience]
What I do all day: [the 4-5 activities that actually fill your week]
What I want from you: [e.g. take routine work off my plate, prepare my meetings, draft first versions I then edit]
How I want to work: [e.g. ask before assuming, keep answers short, always show your reasoning]
Language: [the language you want answers in]

One setting to understand before you go further: approval. In the bar under the message box you will see a word like "Manual". That is the approval mode: Claude asks your permission before it acts on your files, every time. Keep it exactly there. Other modes let Claude act without asking; they exist for experienced users running trusted routines, and they are not for week one. For now the rule is simple: Claude proposes, you approve.

Step 4 · Audit your week

The step the first three exist for

This is where it becomes yours. Describe your real week and let Claude go through it hour by hour, marking where it can genuinely take work off your plate. Only tasks you already do. Nothing hypothetical.

Here is my typical working week, roughly hour by hour:
[Monday: ... Tuesday: ... — describe it the way you would to a new assistant. Paste a calendar screenshot instead if that is easier.]

Go through it and mark every task where you could realistically save me time: preparing, drafting, summarizing, checking, chasing, reformatting. For each one tell me: what you would take over, what stays mine, and roughly how much time per week it saves. Rank the list by time saved. Do not suggest anything I don't already do.

Whatever your role, expect the top of the list to be unglamorous: preparing a recurring review, chasing confirmations, turning a messy export into five readable lines. Nothing on it will say "AI strategy", and that is exactly right. AI is an enabler of the job you already have, not a new job.

Step 5 · Pick one thing and make it repeat

Where this becomes a habit, not an experiment

Take the top item from your audit. Not three items, one. This is your assignment for the coming week:

1. Today: do the task once with Claude, on real material. Drop the actual files into your test folder and say what you want, the way you would brief a capable new colleague. Example: "prepare my Monday demand review from these exports, one page, decisions first."
2. Same session: correct it. The first version will be 70% right. Say what is wrong in plain words ("too long, wrong emphasis, my boss reads only the numbers"). The corrections are the valuable part.
3. When it lands: tell Claude: "we will do this every week, remember exactly how I like it." It saves your preferences as standing instructions in the folder.
4. Next week: run it again with one sentence: "same as last week, here are the new files." If it holds, you have your first repeating system. That moment, not the download, is when you started using AI.
5. Only now, if the task needs it: connectors: giving Claude access to your calendar or email. They are useful precisely here, when a specific recurring task needs them, and never on day one.

Step 6 · Write prompts that work

The same engine, completely different results

By now you have noticed something: the quality of the answer is the quality of the brief. The same model produces a mediocre draft or an excellent one depending on how you asked. This is the one writing skill worth learning, and it has exactly five parts.

1. Role — who should Claude be for this task? "You are a customer success lead who has run renewal reviews for years."
2. Context — what does it need to know? Your situation, your audience, what the output is for. This is the part most people skip, and the reason most prompts fail.
3. Task — what exactly do you want done? One clear job, not three.
4. Constraints — the rules: what to include, what to skip, what is off-limits.
5. Output format — what the answer should look like: a table, five bullets, one page, decisions first.

You are a customer success lead who has run renewal reviews for years (role). I manage 40 mid-market accounts, and my VP wants a shortlist of the accounts most likely to churn this quarter, based on usage decline and support ticket volume rather than gut feel (context). Review the attached account summaries and identify the 3 accounts at highest risk (task). Explain each choice in 2 sentences, and skip any account renewing after month 9 (constraint). Present the 3 as a numbered list: account name, renewal date, risk reason (output format).

To make it easy: I built a free tool that assembles all five parts for you. Answer its questions about your task and it hands you the finished prompt, ready to paste into Claude. Your answers never leave your browser.

You are running

One folder, one audit, one repeating task, five-part prompts

That is the whole method: Claude knows who you are, you know where it saves you hours, one real task repeats weekly, and you know how to brief the machine. Everything else (more tasks, connectors, deeper features) grows from here at your own pace.