Put Claude Code to work.

§ 00 · Read this first

Hi. This guide is for people who run things — founders, operators, team leads, and anyone with too much on their plate. You don’t need to be technical. You don’t need to write code. If you can write an email, you can do everything in here.

Here’s the promise: by the end, three boring jobs will run every week without you. That’s it. Every page in this guide serves that one goal.

One more thing before we start. AI tools change fast — what’s true today gets better next month. So this guide is a living thing: we keep updating it as Claude Code grows. The newest version always lives right here, at verikal.ai/en/guides/claude-code. Come back and visit.

The goal: three working automations in thirty days. Gently.

§ 01 · What Claude Code actually is

Think of Claude Code as a very smart, very eager new helper who lives inside your computer. You talk to it in plain language — “tidy up this folder”, “turn these invoices into a spreadsheet” — and it actually does the thing. It reads files, writes files, runs little programs, and checks its own work.

A chatbot answers questions. Claude Code finishes jobs. That difference is the whole reason this guide exists.

It’s made by Anthropic, works on Mac, Windows, and Linux, and runs with a regular Claude subscription. And despite the name — it’s not just for programmers. Promise.

§ 02 · It’s not just for code

Most of what a business runs on is files: spreadsheets, reports, drafts, exports, folders full of stuff someone has to deal with. That’s exactly what Claude Code is good at. A few real examples:

• Turn a folder of invoices into one clean spreadsheet
• Draft the weekly report straight from your analytics export
• Clean, dedupe, and reformat a CRM export
• Write and maintain internal docs that are always current
• Build the small internal tool nobody had time to build

Rule of thumb: if the job starts with files and ends with files, Claude Code can probably do it.

§ 03 · Step one: download. Step two: breathe.

Here’s the whole first day. Read it once and notice how small each step is.

1. Download Node.js from nodejs.org — click the big green button, install it like any app
2. Open the terminal (the black window — it’s friendlier than it looks) and paste the one line below
3. Breathe. You just did the hardest part.
4. Go to a folder with real work in it and type claude
5. Say hi. Then ask: “Look around this folder and tell me what’s in it.”

npm install -g @anthropic-ai/claude-code

If the word “terminal” made your shoulders tense up — that’s normal, and it passes in about ten minutes. You only ever type two things: claude to start, and plain English after that. Start in a folder that matters, with real files. Watching it sort your actual mess is where the first “whoa” happens.

§ 04 · The two things that matter most

If you remember only two words from this whole guide, make them these: skills and connections. Everything else is detail.

Think of any good helper, human or not. How useful they are comes down to two things: what they know how to do the way you like it, and which doors they can open. For Claude Code, those two things have names.

• Skills — the things you teach it once, so it does them your way forever
• Connections — the tools and places you plug it into, so it can reach your real work

Teach it skills. Give it connections. That’s the whole game. The next two sections explain each one.

§ 05 · Skills: teach it once, keep it forever

A skill is a little recipe card. You do a job with Claude once — say, turning your monthly numbers into a report — and then you write down how you like it done: the format, the tone, what to double-check. Save that as a skill, and from then on it’s one short command.

You don’t even have to write the recipe yourself. When Claude finishes a job nicely, just say: “Turn what we just did into a skill, so we do it the same way next time.” It writes its own recipe card.

Over a few weeks you build a little recipe box: the invoice skill, the weekly-report skill, the clean-up-the-CRM-export skill. That box is where the value compounds — every skill is a job you never have to explain again.

Every job you do twice deserves a skill.

§ 06 · Connections: keys to the right doors

Fresh out of the box, Claude Code can see one thing: the folder you started it in. Connections change that. A connection plugs it into a tool you already use — Google Drive, Slack, your calendar, your database, your CRM — so it can fetch and update things there by itself.

Under the hood this runs on an open standard called MCP — you don’t need to remember that. Adding a connection is one command:

claude mcp add

Think of connections like keys. A helper with the key to the mailroom can fetch the mail. A helper with keys to every room is more than you need on day one. Hand out keys one at a time, to the rooms where the work actually is.

Start with one connection. Add the next when the first one feels boring.

§ 07 · Pick your first three jobs

Don’t start with the hardest thing in the company. Start with the most boring thing. Make a list of every job that repeats, then run each one through three little tests:

1. Repetitive — it happens weekly or more often
2. File-shaped — the inputs and outputs are files, text, or data
3. Checkable — you can tell if the result is right in under five minutes

Take the top three. Three, not ten — three keeps things calm and clear. Split any big goal into about three parts, and keep splitting until every piece fits in one sitting.

§ 08 · Write the house rules

Claude Code reads a file called CLAUDE.md in every folder it works in. That’s where you write the house rules, once: what this folder is, what to never touch, how to check the work. Write it like a note to a new helper on their first day — because that’s exactly what it is.

And keep the permission questions on. Claude Code asks before it changes files or runs commands. For the first month, answer every one, like teaching someone the ropes. You’re not slowing it down — you’re learning what’s safe to hand over.

• Secrets stay in environment variables — never in the chat
• Give the helper its own accounts and keys where possible
• Start it in the folders it needs, not the root of your machine

§ 09 · From one seat to a team

Once your helper is doing real work, a funny thing happens: the bottleneck moves. The hard part is no longer doing the work — it’s keeping track of who’s doing what, human or AI, and where everything stands.

Helpers need what employees need: a task list, context that survives between sessions, and a place to hand work off. Write things down, keep the context with the job, and let anyone — you, a teammate, or the AI — pick up the next open piece.

This is also the point where some owners decide they’d rather not be the one holding the wrench. Fair. Building and running exactly these automations for small and mid-sized businesses is what we do at Verikal — if you’d rather have month one done for you, book a free 30-minute call and we’ll map your first three jobs together.

§ 10 · Seven friendly warnings

1. Vague asks. “Make it better” gets you a shrug. Say what done looks like.
2. Starting with the hardest process in the company. Start with the most boring one.
3. Skipping CLAUDE.md — and re-explaining your business every single morning.
4. Letting it run unsupervised in month one. Watch first, trust later.
5. Pasting passwords or keys into the chat. Never — they live in environment variables.
6. One giant task instead of small pieces. Split by about three.
7. Giving up after one clumsy answer. Correct it like you’d correct a new hire — it learns your ways surprisingly fast.

§ 11 · The thirty-day plan

• Week 1 — Install it. One real folder. Ten small jobs. Note what felt good.
• Week 2 — Pick your three jobs. Write CLAUDE.md. Run each one end to end, checking everything.
• Week 3 — Add your first connection. Turn your favorite job into a skill. Put one job on a schedule.
• Week 4 — Write down the playbook. Invite one teammate. Pick the next three jobs.

Definition of done: three jobs that run every week without you.

§ 12 · The fine print

Costs: a Claude Pro subscription is enough to start; heavier daily use fits Max; scheduled jobs run on the API, where you pay per use. Begin on Pro, upgrade when your helper earns it.

Security: the helper asks before acting, you approve every step until you decide otherwise, and data sent through the commercial API isn’t used to train models by default. Keep permissions tight and secrets out of the chat, and you’re ahead of most.

And remember — this guide is alive. The tools change, and we update it as they do. The newest version always lives at verikal.ai/en/guides/claude-code.