If you’ve heard people say they “built an app over the weekend” with AI and assumed they were exaggerating or secretly engineers, this post is for you. The tool many of them used is Claude Code, and the good news is that the people who get the most out of it aren’t necessarily the best programmers.

What Claude Code actually is

Think of Claude Code as a junior employee who can read your entire filing cabinet, do the work, and show you the result — except the “filing cabinet” is a software project. It’s an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows, all through natural language commands.

The word “agentic” matters. A spell‑checker waits for you to type and suggests the next word. Claude Code works more like a contractor you hand a goal to. It reads a codebase, plans a sequence of actions, executes them using real development tools, evaluates the result, and adjusts its approach. The developer sets the objective and retains control over what gets committed, but the execution loop runs independently.

You don’t have to take the training wheels off, either. By default it is cautious — Claude Code asks before making changes to your files or running commands. Developers control how much autonomy it has, from approving every action to letting it run more freely. And it isn’t only for engineers: Anthropic reports that product managers, founders, and operations teams have started building working tools by describing outcomes in plain language.

Habit 1: Know the problem better than the tool does

This is the single most counterintuitive finding, and it should reassure you. Anthropic studied roughly 400,000 sessions and found that success is determined by how well a person understands the problem they are trying to solve, not whether they’re trained in coding. In fact, sessions rated “expert” reached verified success more than twice as often as those rated “novice,” and when a session hit trouble, novices abandoned it at several times the rate of everyone else.

The lesson: a clinic manager who deeply understands patient scheduling will out‑build a generic coder on a scheduling tool. Bring your domain knowledge, and be specific. “Make it better” gets you nothing; “flag any appointment booked less than 24 hours out so the front desk can confirm it” gets you something useful.

Habit 2: Write a CLAUDE.md so you don’t repeat yourself

CLAUDE.md is a markdown file at the root of your project that gives Claude persistent instructions — build commands, coding standards, conventions — and it reads this file at the start of every session. Treat it like the one‑page brief you’d hand a new hire so you’re not re‑explaining the basics every morning. The /init command generates a starter version based on your project structure, though the output tends to be bloated — if you can’t explain why a line is there, delete it. A practical rule of thumb is to keep the file small, under about 200 lines, and put your important rules in it, like never committing secrets.

Habit 3: Make it plan before it builds

For anything beyond a tiny tweak, ask Claude to lay out a plan first and read it before approving. Plan mode is worth the few extra minutes upfront for multi‑file changes, unfamiliar projects, and architectural decisions — it prevents Claude from spending 20 minutes confidently solving the wrong problem entirely. It’s the equivalent of approving the blueprint before the contractor pours concrete.

Habit 4: Keep its “desk” clean

Like a person, Claude works worse when its workspace is cluttered with unrelated information. Anthropic’s own advice: if you start one task, ask something unrelated, then go back — context is full of irrelevant information, so use /clear between unrelated tasks. And when you’re stuck in a loop: after two failed corrections, clear and rewrite, because the context is polluted with failed approaches.

Habit 5: Connect it to your real tools — carefully

Claude Code can reach beyond code. Using a connection standard called MCP, it can read your design docs in Google Drive, update tickets in Jira, or pull data from Slack. Useful, but treat access the way you’d treat handing someone your keys: grant only what a given job needs.

Is this hype?

Some of it, sure — but the adoption is real. Anthropic reports that the share of GitHub projects with coding‑agent activity has more than doubled since late 2025, and Claude Code users now spend an average of 20 hours per week using the tool. Most strikingly, at Anthropic itself, the majority of code is now written by Claude Code, with engineers focusing on architecture and direction.

Your next step

Pick one small, annoying task you understand cold — a weekly report you assemble by hand, a spreadsheet you keep reformatting. Claude Code is available for macOS, Linux, and Windows. Install it, write a five‑line CLAUDE.md describing your task, ask it to plan first, and approve each step. You’ll learn more from one real attempt than from a month of reading — including whether your understanding of the problem is as solid as you think.