TL;DR
Last month, I shared with product and business colleagues how to get started with AI Coding. Combined with a recent tweet about how many students miss out on top-tier AI Coding tools due to subscription hurdles, I realized that methods and habits can be learned for free first. I've organized the getting-started portion here. To make it easier to understand, I've included several simple illustrations to make the concepts more direct.
Many people get stuck at the first step of using Claude Code: the command line. Seeing a terminal with only text makes them think it's only for programmers. In reality, the threshold isn't that high. Anyone who can use a chat AI like Doubao can get started with a little time; the rest is just getting used to handing over execution to the AI.
Once you get the hang of it, you'll find it's like a capable assistant that takes on any task: running backend data, writing small tools to solve your problems, merging messy documents into briefings, making prototypes, or organizing sales reports. Whether you could write code before isn't the key. Once you consciously write project context into CLAUDE.md, make your requirements precise, and think about creating "Skills" to package repetitive actions, you've essentially entered the door. This article is meant to lead non-technical students into using my favorite tool, Claude Code.
The First Hurdle: The Command Line
Non-coding students are used to chat AIs like Doubao. The first time they install Claude Code, it feels a bit unfamiliar. Previously, it was a back-and-forth process: you describe the need, it generates code, and you copy-paste it elsewhere to test. Now, Claude Code runs directly in the terminal, skipping the manual transport step.
If you've never used a terminal, I recommend Kaku, which I created specifically for AI Coding. It's ready to use out of the box without messing with colors or fonts. It follows system light/dark modes, uses Cmd + D for split screens, and Cmd + Shift + Y for the file manager. Most importantly for beginners, it has built-in AI assistance: it suggests fixes for command errors and can generate commands if you type # followed by Chinese.
Installing Claude Code takes just one command (see official docs), then just type claude in your project folder to start coding.
curl -fsSL
| bash
Highly Recommended: Brush Up on Technical Knowledge
To truly use Claude Code well, describing needs isn't enough. Understanding basic concepts makes troubleshooting much easier.
Know what common frameworks do: Understanding what React, Vue, or Next.js solve helps you not feel lost when looking at Claude Code's output.
Software basics: Terminal commands, Git, VS Code, and Chrome DevTools. When an error occurs, you can help locate it rather than just waiting.
Core programming ideas: What functions are, what variables and states are, and why code is split into multiple files. This makes your requirements more precise.
Learn to read code and errors: This is useful even before you can write code. Glancing at a change to see what it does is faster than asking for an explanation. Don't panic at errors; copy the whole thing back and ask "What does this mean and how do I fix it?" Nine times out of ten, it will tell you exactly which line is the problem.
You don't need to reach the level of writing code yourself; just knowing what these things look like is enough. Spend an evening or two on freeCodeCamp or MDN's intro sections, or watch an intro course like Crash Course Computer Science or Harvard CS50. It will significantly change your collaboration efficiency.
I recommend these three books for a broader perspective: "Inspired" for product judgment, "The Linux/Unix Design Philosophy" for engineering philosophy, and "Listening to the Wind" for an expert programmer's vision.
Preparation: Accounts and Subscriptions
Account: Register at claude.ai using Gmail. Use a stable US IP to avoid risk control triggers. Do not jump straight to a Max subscription on a new account to avoid bans.
The easiest way is via the US App Store or Google Play. Subscribe to Pro in the Claude App using your balance. Note that App Store has taxes (a 100 unit sub might show as 125), but it's safer. Start with Pro and upgrade to Max if the quota isn't enough.
To avoid bans: use a stable network, don't share accounts, use reliable payment methods (avoid virtual cards), and use an established Gmail account.
What Tasks Suit Claude Code?
I've tried Cursor, Windsurf, and Codex, but Claude Code is my daily driver.
Its model capability is excellent, and its implementation of the "Harness" concept is top-tier. It scans CLAUDE.md and the directory structure to understand context, then modifies files, runs commands, checks errors, and iterates on its own. Since it lives in the terminal, it can directly call git, tests, and scripts.
Claude Code is actually a general-purpose Agent. Anthropic's internal sales and finance teams use it to process CRM data and emails. If you hate the terminal, use the desktop app Cowork, which can read/write your local directories to do things like turning receipt screenshots into expense reports.
Crucially: in coding, accuracy is more important than speed. It's better if it takes 20 minutes and works than if it takes 10 minutes and requires 20 minutes of debugging.
Best tasks: Prototypes and internal tools, processing CSVs/sales reports, summarizing contract clauses, and extracting info from PDFs/links.
Build Software for One
NYT columnist Kevin Roose proposed the concept of software for one: you don't need an app for a million people; you can make software just for yourself. This could be a tool to turn voice notes into meeting minutes or a personal dashboard. Product and business people are best at this because they know their own pain points.
The 1-Day to 3-Month Rhythm
Don't try to build Notion on day one. Follow this rhythm:
Day 1: Modify an existing Excel or Markdown file. Week 1: Build a single-page personal homepage. Month 1: Automate a repetitive weekly task. Month 3: Build a "software for one" tool.
Turn Web Actions into Commands with OpenCLI
Operations staff often spend all day clicking in browsers. These tasks can often bypass the UI via APIs.
OpenCLI provides CLI adapters for Xiaohongshu, X, Bilibili, etc. It allows Claude Code to perform web actions like clicking, inputting, and scraping with a single command.
Xiaohongshu Research: Have Claude Code use OpenCLI to scrape data and summarize hot keywords.
Sentiment Summary: Combine adapters for X, Reddit, and HackerNews into a daily report.
Claude Code also has a Routines feature to automate workflows via Webhooks or GitHub events.
CLAUDE.md: Define Project Context First
Many people get frustrated because they have to repeat context. The solution is a CLAUDE.md file in the root directory.
Claude Code reads this every time it starts. Keep it under 150 lines, use imperative language, and provide verifiable rules (e.g., "functions must be under 50 lines").
Key rules: Ask before acting, prioritize simplicity, only touch what's necessary, and verify after completion.
Precise Requirements Prevent Deviations
Vague: Help me make a customer follow-up tool.
Precise: Help me make a sales follow-up tool as a single-file local webpage. The left list shows company name, next follow-up, and status. The right side shows details. Add three filters at the top. Store data in localStorage.
Precise requirements get a working version on day one. Even if you don't know the code details, you must describe the requirements specifically.
Define the scope: explicitly state what not to do (e.g., "no login, no SEO for now"). Provide numerical acceptance criteria (e.g., "first screen loads in 1.5s").
Plan and Auto Modes
For complex tasks, use Plan mode (Shift+Tab). It lists what it intends to do before executing. If you're comfortable, use Auto mode, which performs safe operations automatically and only asks for permission on risky ones (like deleting files).
How to Confirm It's Right
Don't just take its word for it.
Git Snapshots: Run git status before big changes and commit checkpoints. Use /rewind to undo changes.
Avoid the "Try and See" Loop: Don't let it patch code without a diagnosis. Ask "Which line is the problem and why?" before allowing changes.
Max Advanced Tips: Aliases, Models, and Long Sessions
Aliases: I use alias c='CLAUDE_CODE_AUTO_COMPACT_WINDOW=400000 claude --dangerously-skip-permissions' to skip confirmations and delay context compression.
Models: Use /model opusplan to let Opus handle planning and Sonnet handle execution.
Long Sessions: Use /clear after finishing a task. Ask for a HANDOFF.md before ending a long task so you can resume in a fresh session.
Waza: Turning Habits into Muscle Memory
I created Waza, a set of 8 skills representing good engineering habits.
/think for planning, /design for UI design, /hunt for troubleshooting, and /check for final reviews.
Create Your Own Skills
Skills are folders in .claude/skills/ with a SKILL.md. They can be workflow-based (e.g., summarizing meetings), checklist-based (e.g., pre-launch checks), or expert-based (e.g., troubleshooting matrices).
Kami and Claude Design
Kami is an AI layout tool to turn content into professional PDFs. Claude Design allows you to create interactive prototypes from screenshots or docs without writing code.
Security Considerations
Have it explain commands before running them. Never use production environments for practice. Don't paste API keys directly into the chat; use .env files. For sensitive features like payments, use established services like Stripe rather than writing from scratch.
