Tourist Guide

Setup Claude Code in 60 minutes

A guide for Tourists ready to become Builders.

You’re not behind. You just haven’t seen the patterns yet.

Right now you’re using a phrasebook in a country where everyone speaks the language. The technology isn’t the problem. Your mental model is.

Most operators in your spot hit limits weekly, walk away convinced AI is overhyped, and never come back. The fact that you’re reading this means you’re not most operators.

So here’s the deal.

The gap between Tourist and Builder is five specific habits. Each one takes minutes to install. None require code. None require engineering background.

You can do all five in an afternoon. The internalization takes about two weeks of practice.

By the end, you’ll be shipping things you couldn’t ship before, with a fraction of the token spend.

Here’s the path.

What you’re actually working with

Quick reset before the install steps. This is the mental model that makes everything else click.

Claude Code is not ChatGPT with files.

It’s an agent. It reads your work. It plans across multiple steps. It executes. It iterates. The execution loop runs independently of you.

That last part matters. Most Tourists treat Claude Code like a chatbot. They type a question. They wait for an answer. They type another question. They wait for another answer.

This is how you burn through plan limits in three days.

Builders treat Claude Code like an agent. They give it context. They give it a task. They let it run. They check the result. They iterate.

The difference isn’t in what they ask. It’s in how they set up the work before they ask anything.

That’s what these five installations do. They set up the work.

Install 1: Learn /clear

This is the most important command in Claude Code. Most Tourists don’t know it exists.

Open Claude Code. Type /clear. Hit enter.

That’s it. Your conversation context resets. Your CLAUDE.md and project files stay. The chat history is gone.

Why this matters: every message in your session gets re-read by the agent on every new turn. A session that’s been running for two days with twenty unrelated questions in it forces the agent to re-process all of that on every new question. The cost compounds invisibly until you hit your limit.

The rule: new task, new session.

You’re not switching to a marketing email after debugging a script in the same chat. You’re running /clear first, then asking the marketing question.

Anthropic’s own documentation calls this the single most effective lever for both quality and cost. They’re not wrong.

Time to install: 30 seconds. When to use it: Every time you switch tasks. Every time you finish something. Every time you’re not sure if your context is bloated.

Install 2: Switch your default model to Sonnet

Your default model is probably Opus. That’s what gets recommended when you sign up. It’s the most powerful model. It’s also the most expensive.

Opus costs roughly five times what Sonnet does in plan tokens. For most tasks, the quality difference is invisible.

Type /model in Claude Code. Pick sonnet-4.6.

You just got back roughly half your plan capacity. No quality loss on routine work.

Reserve Opus for things that genuinely need it. Complex architectural reasoning. Multi-file refactoring with subtle dependencies. Hard debugging that requires holding a lot of state in mind.

For everything else, Sonnet handles it.

Even better: when you hit a complex task, use /model opusplan. This uses Opus to plan and Sonnet to execute. You get Opus-quality reasoning where it matters and Sonnet pricing on the bulk of the work.

Time to install: 30 seconds. When to use it: Right now. Then leave it alone.

Install 3: Create a CLAUDE.md

CLAUDE.md is a file that lives in your project. The agent reads it at the start of every session. It’s how you give Claude Code persistent memory about what you’re working on.

Without one, every session starts blind. You spend the first twenty minutes re-explaining your stack, your preferences, your conventions, and what you were doing yesterday.

With one, every session starts smart.

Create a file called CLAUDE.md in your project directory. Drop in:

# Project context

What this project is:
[One paragraph]

Key conventions:
- [Bullet point]
- [Bullet point]

Tools and stack:
- [List]

Things to avoid:
- [List]

Keep it under 200 lines. If you don’t have anything for a section, leave it blank. This is a working document, not a deliverable.

Update it whenever the agent makes the same mistake twice. First time is a one-off. Second time is a pattern. Patterns get rules.

After three months of this discipline, your CLAUDE.md becomes a codified version of every lesson you’ve learned. The agent stops making whole categories of mistakes you used to catch manually.

Time to install: 15 minutes for the first version. Three minutes per update. When to use it: Read it back to yourself once a week. Cut anything that’s stale.

Install 4: Scope your prompts

Most Tourist prompts look like this: “Analyze my repo and tell me what’s wrong.”

That prompt forces the agent to grep dozens of files, read them partially, produce a long summary, and burn through a quarter of your daily allotment in a single request.

The Builder version of the same goal: “Look at the API authentication code in /auth/login.py. Check for security issues related to session handling. Output a numbered list of findings with file and line numbers.”

Same goal. A third of the tokens. Better answer.

The pattern: specify the file or directory, specify the type of issue you’re looking for, specify the format of the output.

You don’t need to write long prompts. You need to write specific ones.

If you can’t specify what you’re looking for, you don’t actually know what you want yet. Stop and figure that out before burning tokens on a sweep.

Time to install: No installation. This is a habit you build by catching yourself before every prompt. When to use it: Every prompt. Forever.

Install 5: One topic per session

You asked Claude Code to help with a marketing email. Then a pricing analysis. Then debugging a script. All in the same session.

Every old message is dead weight that gets re-read on every new turn. Your marketing email is making your debugging slower and more expensive.

The rule: new topic, new session.

Switch from email to pricing? Run /clear. Switch from pricing to debugging? Run /clear. Switch from debugging to documentation? Run /clear.

This sounds aggressive. It’s not. Each /clear takes one second and saves you tokens for the rest of the session.

Pair this with Install 1 and Install 3. CLAUDE.md persists across /clear. Your project context never gets lost. Only the chat history goes.

Time to install: No installation. This is the habit that ties Install 1 to your daily workflow. When to use it: Every time you mentally switch contexts.

The two-week internalization

You can install all five in an afternoon. Internalizing them takes about two weeks.

Here’s a realistic ramp.

Days 1 through 3: Use /clear aggressively. Every time you finish something. Every time you start something. Every time you’re not sure. Build the muscle memory.

Days 4 through 7: Make Sonnet your reflex. Notice when you’re tempted to switch to Opus. Ask yourself if the task actually needs it. Most of the time it doesn’t.

Days 8 through 11: Live in CLAUDE.md. Update it every time the agent makes a mistake twice. Read it back to yourself at the end of each session. Cut what’s stale.

Days 12 through 14: Tighten your prompts. Before every prompt, ask yourself: am I being specific about the file, the issue, and the format? If not, rewrite before sending.

By day fourteen, the five habits are reflex. You stop thinking about them and they just happen.

That’s when you’ve crossed the threshold. You’re not a Tourist anymore.

What changes when you cross

The first thing you’ll notice is that you stop hitting plan limits.

The second thing you’ll notice is that the agent gets better at what you want without you working harder. Your prompts are scoped. Your context is clean. Your CLAUDE.md compounds. The output quality goes up while your token spend goes down.

The third thing you’ll notice is that you start shipping things you couldn’t ship before. Not because Claude Code got better. Because you got better at using it.

That’s Builder fluency.

The leap from Builder to Orchestrator is bigger and takes longer. Twelve weeks instead of two. But you’re not there yet, and you don’t need to be.

Right now, focus on the five installations. Get them into reflex. Then take the assessment again in two weeks and see where you land.

If you’ve made it this far, you have everything you need to start.