Chapter 01 of 8
The idea.
The whole setup rests on one idea: you set the terms and Claude does the work. It sounds obvious, yet most people end up doing the opposite, taking orders from their AI one pasted command at a time.
The whole setup in this guide rests on one idea. You are the boss, and Claude is the crew. You decide what gets done and what "done" means. Claude does the work, all of it, and shows you proof. It sounds obvious. Yet almost everyone who starts with Claude ends up doing the opposite, and never quite notices.
The mistake almost everyone makes.#
Out of the box, Claude acts like a very smart consultant. It explains. It suggests. It hands you commands to paste into a terminal (the text window where programmers type instructions to the computer). It asks which of four options you prefer. It tells you what you "will need to do next."
If you go along with that, you become Claude's assistant. You paste what it says, click what it says, and make every small decision it hands back to you. It feels productive. It isn't. You're doing the work, and the AI is managing you. Flip that around and the whole picture changes: you describe the outcome in a sentence or two, and Claude handles every step between here and there, stopping only for the few things that truly need your hands or your judgment.
What a normal day looks like.#
Here's a normal day once this is set up. You open the Claude app on your laptop, type or talk for thirty seconds, and walk away. While you're gone, Claude builds the page, checks it on a phone-sized screen, fixes what broke, puts it on a preview link, drafts the email to the client, and leaves you a two-line note: what's done, and what it needs from you. A second Claude on a Mac mini in the office runs the scheduled work all night, checking your sites, updating your reports and writing your morning brief.
No developers. No virtual assistant. One person, two computers, and a set of habits. It works the same whether your business is an agency, a store, a practice, or a company that runs on spreadsheets.
Under the hood, it's simpler than it sounds, and the stack below shows how the pieces sit on each other. Your laptop runs the Claude desktop app, pointed at one folder, ~/Code (a folder called Code inside your home folder), that holds every project you have. Behind it sits a brain: a folder of plain-text notes Claude reads at the start of every session, plus one file of standing orders called CLAUDE.md. The brain is backed up to GitHub (an online home for project files, with a full history of every change) and shared between your computers, and every project lives in its own private GitHub repository, so nothing lives only in a chat window. Connectors, switches that each let Claude work inside one app like Gmail, Calendar, Drive or your accounting system, plug in your tools. Meeting transcripts become Claude's brief. And a short list of guardrails spells out what Claude never does without you. The rest of this guide builds each piece, and from Getting set up onward, the computer does most of the building itself.
What sits under a normal day
The pieces this guide builds, from the one you touch down to the machines.
So what do you actually point all this at? Most people hear "Claude Code" and picture software. In practice, the best uses are reports, documents, spreadsheets and small jobs that run on their own. Think about last Monday. How much of it went to copying numbers into a spreadsheet, rewriting the same email, or hunting for a file? For most businesses that's hours a week. Here's what reclaiming them looks like in a few very different kinds of business.
Out in the field and behind the desk.#
Start with a roofing or cleaning company. The first thing most of them build is a quote calculator: the pricing rules that live in the owner's head, or in a messy spreadsheet, turned into one page that works on a phone, so every estimator quotes the same price for the same job. Asking for it takes a sentence or two, and the figure after it shows how the work splits once you press Enter.
Our pricing rules are in pricing.xlsx. Build a one-page quote calculator that works on a phone, test it on five real jobs, and show me a screenshot of each.
Who does what on one job
The quote calculator request, step by step, with the owner of each step.
Next comes the lead flow. Every web form and ad lead gets filed in the CRM (the program where you keep customer contacts and deals) with its source attached, and someone gets an alert while the lead is still warm. Once that's in place, Claude can show what each paying customer actually cost, by ad channel, and which jobs made money after materials, labor and fees. The busiest kind of job is often not the profitable one. And when the owner talks through a job into a voice memo once, Claude turns it into a clear how-to guide with a checklist for the next hire.
A law or accounting office has different paperwork but the same problem. An intake form that files itself is a common first win: each new client's answers land in a spreadsheet, and missing documents get flagged before anyone chases them by phone. From there, Claude drafts proposals from sales call notes in the style of your last three, using the client's own words and leaving the prices blank for you. It can take a first pass at a contract and flag risky clauses against your standards (that speeds up review, but it isn't a lawyer). In tax season, an accounting firm can make a personal letter for every client from one spreadsheet in an afternoon.
Now picture a clinic's front desk. The new-hire checklist, with names and due dates for the first two weeks, becomes a shared list instead of a binder. When a policy changes, Claude finds and fixes every section of the handbook it touches. A year of patient questions becomes a clear page of answers and a set of approved replies, and the inbox gets sorted each morning with routine answers already drafted. A person reads and sends them.
Behind the counter and in the office.#
An online shop usually starts with low-stock alerts: a daily list of what to reorder, based on what actually sold. Then a month of social posts, with on-brand graphics and captions the owner approves in one sitting, and clean product sheets with photos and prices for wholesale buyers. A restaurant can paste in a year of reviews and get back the top themes, with real quotes. And any business with a website can have a nightly check that loads the key pages, tests the contact form with an obviously fake name, and speaks up only when something breaks. Jobs like that one run best on a computer that stays on, which is what The always-on setup is for.
An agency lives on repeat work. The monthly client deck gets built the same clean way every month, straight from each client's numbers, with every figure checked against the spreadsheet. Before a pitch, Claude makes a table of what competitors charge and include, taken from their public websites with the page named for every fact, and a prospect list with a short note on why each company might need you. After every call, the transcript becomes decisions, owners, due dates and a draft follow-up. Working with Claude every day shows how to get those transcripts to Claude.
A nonprofit gets the donor report it never has time for: giving by campaign, plus a list of lapsed donors worth a phone call. The quarterly board slides come straight from the finances. The contact list that's been merged and re-merged for a decade finally gets cleaned, with duplicates removed, nothing deleted, and every change logged for review. And a 13-week cash forecast, built from the bank export and the bills, shows a tight week coming while there's still time to do something about it.
Almost every business ends up wanting the Monday report, too. You drop the weekly exports into a folder, and one page comes back.
This week's exports are in reports/weekly. Make me a one-page Monday report: totals against last week, anything more than 20% off, and three things I should look at. Show the math on a second tab.
A few of these need Claude to see your email or files, which is what connectors are for, and Teaching Claude your business explains which to turn on. Anything that sends messages, moves money or touches private client data gets built the same way, but the final send or pay step stays with you. Staying in charge explains how.
How to pick your first three.#
Don't start with the most impressive idea. Start with the job you do every week and dread every time.
For each idea that fits your business, ask three questions. How often do you do it? Weekly jobs pay back fastest. How much do you dread it? You'll stick with the ones that annoy you. And how safe is a bad first try? A draft or an internal report can be wrong without hurting anyone, while anything that touches money or customers should wait until you trust the setup. The matrix below sorts this chapter's examples on two of those questions, and for most businesses the winners land in the same corner: a weekly report, a spreadsheet cleanup, a document you rewrite often.
How to pick your first jobs
Examples from this chapter, sorted by how often they come up and how safe a bad first try is.
New leads into the CRM
Low-stock reorder list
Call notes to follow-ups
Tax-season client letters
Handbook policy fixes
A year of review themes
Then brief Claude like a new hire: the result you want, where the files are, and what it must not touch, and ask for proof before you rely on it. Once a job runs well, ask Claude to save the steps as a skill, a small folder of instructions (kept under ~/.claude/skills) that Claude picks up on its own whenever a task matches. Next week you ask for it in one sentence, and the list of things you do by hand gets a little shorter.
Why letting Claude steer makes the work worse.#
Every one of those jobs goes better when you hold the wheel, and not just because your time is worth more. Letting Claude steer makes the work itself worse, in ways that are easy to miss. Left alone, Claude drifts toward the safe, generic answer: the most common design, the most common wording, the most common tool. Your business isn't common, so you have to keep steering it back to what you actually want. Every handoff you accept becomes the pattern, too. Run one command it gave you and it will give you the next one. Answer one "which do you prefer?" and ten more follow. Push back once, clearly, and it adjusts for the rest of the session.
Then there's the word "done." Claude will sometimes say something works when it hasn't checked, and if you don't ask for proof, you find the bug later, usually in front of a customer. And it forgets, unless you make it remember. Every correction you give can be saved as a permanent rule, one line in CLAUDE.md or a short note in its memory folder, which Teaching Claude your business covers. People who skip this correct the same mistake forever. People who do it get a Claude that gets sharper every week.
Put those pieces together and you get the simple loop drawn below. You ask for the outcome, not the steps. Claude does all of it and proves it worked. You push back where it went wrong, and that correction gets saved so the next trip around starts from a better place.
Push back is not the same as ignore.#
None of this means you tune Claude out. When it flags a real risk (sending an email, spending money, deleting something, a legal or security problem), stop and read it. Claude is often right about risk. It is usually wrong about whose job the work is. The trick is keeping those two apart.
| Listen when Claude says | Push back when Claude says |
|---|---|
| "This will send an email to 400 customers." | "Run this command in your terminal." |
| "This deletes the old data and it can't be undone." | "Which of these four options do you prefer?" |
| "This API key is about to be committed to GitHub." | "You'll need to manually go into settings and..." |
| "This costs about $50 in API credits. Proceed?" | "I can't access that." |
| "The client said X on the call, but this does Y." | "Done! Everything should work now." |
Staying in charge has the full pushback playbook, with the exact words to use, so read it before you start. Everything else in this guide comes back to one rule, and it's worth keeping somewhere you'll see it every day.
Don't do what Claude tells you to do. Tell Claude what to do, and make it do all of it.
Common questions.
What is Claude Code?
Claude Code is Anthropic's AI helper that works directly on your computer. It can read and change files, run programs, and build tools, reports and documents. Most owners use it in the Code tab of the Claude app.
Can I use Claude Code if I don't code?
Yes. You describe the result in plain English, and Claude writes and runs any code itself. Your job is to judge the result, approve risky steps, and push back when it hands work back to you.
Is Claude better than ChatGPT for business?
Neither is best at everything, and both change every month. Claude is strong at long documents, careful writing, and doing work on your own computer. The best test is to give both the same real task and compare what comes back.
What can Claude Code do for a business?
It can build small tools, connect your apps, make reports, clean spreadsheets, draft documents and slides, and research your market. It works on your own computer with your files. You describe the result and check what comes back.
What are some Claude Code examples for non-programmers?
A weekly sales report built from exports, a cleaned-up customer list, a proposal drafted from a call, and a monthly client slide deck. None of them need you to read or write code.