From Claude chat
to Claude Code.
In 20 minutes.
You're going to learn how to use Claude in your browser, then climb the 2 steps above it — all the way to Claude Code, without ever writing a single line of code. You speak in plain English, Claude does the work. It's that simple.
The advice of Ethan Mollick (professor at Wharton, author of Co-Intelligence). You'll learn more by using Claude for 10 hours on tasks you already know than by taking 50 prompt engineering courses. — Ethan Mollick · One Useful Thing
What you'll walk away with
- Ten hours of use beat fifty courses. Use Claude on tasks you already know, before you go hunting for the magic trick.
- Claude has 3 doors: the Chat (your browser), Claude Cowork (on your computer) and Claude Code (your personal workshop, without writing a line of code).
- Yes, Claude Code is for you too. We call it "Code" because of where it came from, but you talk to it in plain English. I've never coded a day in my life, and I use it every day.
The 7 words to know before you start.
If you read articles about AI and people throw these words at you without defining them, you lose the thread. Here are the 7 terms that come up all the time, translated into plain English.
| Word | What it means |
|---|---|
| Anthropic | The company that builds Claude (the way OpenAI builds ChatGPT). |
| Claude | The AI itself. What answers you when you ask a question. |
| A prompt | The text you send to Claude. Your question, your request, your instruction. |
| Pro | The paid Claude plan at $20 a month. Gives you access to the most powerful models and to Cowork. |
| Projects | A feature inside Claude.ai: a folder where Claude keeps your context in memory from one conversation to the next. |
| Cowork | A Claude assistant that runs on your computer and can act on your files, your emails, your Drive. |
| Claude Code | The most powerful version of Claude, running directly on your computer. Despite the name, you don't need to know how to code — you talk to it in plain English, and it acts on your files. |
The 3 doors into Claude.
Before you choose, you need to know they exist. A lot of people get stuck on the first one because they've never heard of the other two. Picture a staircase: hotel room → shared office → your own workshop. Each step is more powerful and more personal.
| # | Official name | What is it? | Who's it for? | Price |
|---|---|---|---|---|
| 1 | Claude (the Chat) | The Claude.ai app in your browser — like ChatGPT, the Anthropic version. | Everyone, starting today | Free / $20 a month (Pro) |
| 2 | Claude Cowork | An assistant that runs on your computer (Mac or Windows) and can read, write, and organize your files, send emails — no coding. | When you want to hand off a whole task, not just ask a question | Included in Claude Pro ($20 a month) |
| 3 | Claude Code | Your personal workshop on your computer. You speak in plain English, it acts on your files, it remembers between sessions, it can work while you sleep. | For you too — as long as you do steps 1 and 2 first | Included in Claude Pro |
Door 1 · The Claude Chat
The metaphor: it's the hotel room. Clean, ready to go, you walk in, you ask your question, you walk out. Nothing is kept for next time.
You go to claude.ai, you create an account, you type your question in the bar at the bottom. That's it. Exactly like ChatGPT, except it's Claude — an AI known for being more precise at writing and more honest when it doesn't know.
Real example: you paste in a tricky client email and ask "reply to this in a polite but firm tone, in 5 lines." You've got your answer in 10 seconds.
Tip: to keep your context from one conversation to the next (your offer, your tone of voice, your audience), use the Projects feature in Claude.ai Pro. It's a folder where you paste your documents once, and every new conversation starts with all that context in mind.
Door 2 · Claude Cowork
The metaphor: it's the shared office. Claude now has access to your stuff. It can read your files, reorganize them, help you on a concrete project. But you stay right next to it to approve every step.
Cowork is software you install on your computer. Instead of opening a web tab, you talk to Claude in a dedicated window — and this time, it can act on your computer: read your files, rewrite them, sort a folder, help you organize emails.
Real example: you tell it "take the 30 customer reviews in this file, sort them by theme, write me 30 replies in my usual tone, put it all in a new document." You let it work, you come back, it's done.
This is the right door the moment you want to hand off a whole task, not just ask a question. And you don't need to know how to code.
Door 3 · Claude Code
The metaphor: it's your personal workshop. You arrange it your way, you leave your tools in it, you write your rules on the wall. Claude walks into that workshop, reads the rules, works the way you want. And the next day, it remembers everything.
Despite the "Code" in the name, you don't need to know how to code to use it. You talk to it in plain English, it does the work. Two ways to launch it:
- The window app (Mac/Windows) — no command to type. You click an icon, you write like you would to a coworker. It's the most accessible version to start with.
- The terminal — a text window on your computer. 3 commands are enough to install and open it. Nothing else to learn.
Real example: "organize my Downloads folder by file type, archive anything older than 6 months into an Archive folder." You approve, it's done. Same thing for writing an article from your notes, building an expense report from a bank export, turning a meeting into a structured recap.
A bonus worth knowing: the moment Claude Code is running on your machine, you can install official Anthropic plugins. They're like apps for your phone · one command and Claude gains a new superpower (up-to-date documentation, a dev methodology, an automatic audit of your rules). Don't stress about this at the start · it'll come naturally when you feel what you're missing.
Second bonus: GitHub · the vault where Claude saves your work. No need to know how to code to use it · 5 commands are enough, and you can ask Claude to run them for you. The moment you build something you don't want to lose, set it up.
My full step-by-step guide shows you how to install Claude Code in 10 minutes, then my GitHub page walks you through the safety net.
The staircase I recommend
Start with the Chat (door 1) today. When you notice you're re-pasting the same context several times a week, turn on Projects in Claude.ai Pro. And when you feel like you want Claude to act (organize, write into files, send emails), open door 3. You'll find there's nothing scary about it — just a bigger, more personal workshop.
Claude, a brilliant intern with no memory.
It's the best analogy I've found, and it comes from Ethan Mollick. If you really take it in, you avoid 90% of the disappointments beginners have with AI.
Claude has read half the internet. It can reason, write, analyze, code. But it has two limits you need to understand so you're never let down.
It forgets everything between conversations
With every new conversation, it's an intern showing up at 9 a.m. with zero context. You have to brief it: who you are, what you do, what you expect, for whom. Without that, it hands you something generic.
It doesn't know what it doesn't know
It can invent a statistic, cite a study that doesn't exist, be wrong with total confidence. Your job: check the facts it gives you, especially numbers and quotes. A number that's too round or a source that's too perfect? Google it before you use it.
The right reflex
Treat Claude like a talented new hire who needs onboarding every single time. The more context you give it, the better it is. The less you give it, the more it makes up. It's that simple.
The #1 setting that 90% of users ignore.
If there's only one technical tip you take from this page, this is it. There's a spot in Claude.ai where you can write once and for all who you are and how you want Claude to talk to you. And it'll remember it in all your conversations.
Where to find this setting
In Claude.ai, click your name in the bottom left, then Settings → Personalization. You get a free-text field. That's where it all happens.
The template to adapt
Here's a base you can copy-paste into that field. Replace the brackets with your own information.
My name is [First name]. I'm a [job, industry, size of organization].
My clients are [type of clients].
My communication tone: [3 adjectives, for example: direct, warm, precise].
When I ask you for something:
- Answer in a structured way (lists, tables when useful)
- No courtesy compliments ("Great question!")
- No disclaimers ("I'm only an AI", "It's important to note…")
- Cite your sources when you give a number
- If you're not sure, say so clearly
What I do NOT want you to do:
- [examples: "no bullet points in my emails", "no emoji", "no buzzwords"]
Why it changes everything
Every one of your conversations starts ten levels higher. No more re-pasting your context each time, no more reminding it of your tone. It's the setting with the biggest impact on your quality of use, and almost nobody uses it.
↗ The big brother: the CLAUDE.md
These Custom Instructions are the web equivalent of the CLAUDE.md file that exists in Claude Code. Same logic, more powerful: you write your rules once, Claude Code reads them at every session, and you can have a different one per project. I share mine, anonymized — download it, adapt it to your stack, keep it handy for the day you move up to Claude Code.
The CRAFT method.
A prompt is just the text you send to Claude. But between a sloppy prompt ("write me an email") and a prompt that works, there are five elements. The CRAFT method is a summary of what Anthropic and OpenAI recommend in their own guides.
Who you are, who you work for, what your constraints are. Without it, Claude answers for nobody.
"You are [specific expert]." It activates the right expertise in its memory.
A precise verb: analyze, compare, rewrite, list, fix. Not "make me a proposal."
List, table, 200 words, 3 paragraphs, markdown. Otherwise you get back an unreadable wall of text.
One or two examples of what you expect. It's the most powerful lever of the five.
5 job-specific prompts, ready to copy
There's no universal prompt. Here's one prompt per profile, built with CRAFT, that you can paste into Claude right now. Replace the brackets with your own info.
You are my strategic right hand, with the eye of a senior consultant. Here's my offer: [describe in 3 lines] Here's my ideal client: [describe in 3 lines] Here are my last 3 months of revenue or activity: [numbers] Identify: 1. The #1 bottleneck that's costing me revenue 2. 3 80/20 levers I can pull this week 3. The business risk I'm not seeing Be direct. No courtesy compliments.
You are an expert in online reputation for the restaurant business. Here are 30 Google reviews of my restaurant [name, type, city]: [paste your reviews here] Give me: 1. The 3 recurring complaints to fix in the kitchen or in service 2. The 3 strengths to amplify in my communication 3. 10 template replies (5 positive, 5 negative) in the voice of the house: [describe your tone in 3 lines] Format: a clear table, no filler.
You are a copywriter specialized in direct-to-consumer e-commerce. Here's my current product page: [paste] Here are 20 recent customer reviews: [paste] Here are 2 competitors who are performing well: [URLs] Rewrite the page in 9 sections: hook, unique proposition, benefits, objections, social proof, FAQ, guarantee, call to action, micro-copy. Use my customers' exact words from the reviews. Brand tone: [3-line description].
Act as a senior consultant at a tier-1 firm. Strictly apply: - The Minto Pyramid (conclusion first) - SCQA (Situation, Complication, Question, Answer) - MECE for any breakdown - "So what?" on every point Engagement context: [5-line brief] Data to analyze: [paste] Expected deliverable: a one-page executive summary + 5 prioritized bullets. Be hypothesis-driven, not descriptive. IMPORTANT: I anonymized the client names before pasting. You can work with peace of mind.
You are a senior Creative Director at an independent agency. Client brief: [paste the brief] Audience: [personas in 3 lines] Brand (tone, values, no-gos): [describe] Deliverable: 1. Big idea in one sentence 2. 3 creative directions (tagline + visual description + tone of voice) 3. Adaptation into 6 formats (Instagram reel, story, LinkedIn, TikTok, print, out-of-home) 4. Metrics to track No-gos: clichés, easy puns, the words "unlock", "elevate", "empower".
The tip that changes everything
When you find a prompt that works for you, save it in a dedicated file. A good prompt reused 50 times is worth more than a brilliant prompt used once.
Your first real test, right now.
You've got the prompts, you've got the method. All that's left is to actually do it. If you've never created a Claude account, here are the 6 steps. Count on 10 minutes, no more.
Open a tab on claude.ai
Type claude.ai in your browser. You land on Anthropic's home page.
Create a free account
Click Sign up in the top right. You can log in with Google in one click, or create an account with your email. It's free.
Come back here and copy the prompt that matches your job
Go back to the section just above. Click Copy in the box for your profile. The prompt is on your clipboard.
Replace the brackets with your real content
If you don't have an About page handy, grab your LinkedIn profile, an intro email, any text of your own. It doesn't matter, as long as it's about what you do.
Paste it into Claude and hit send
You've got your first answer. Read it. If it's generic, that's normal: you probably didn't give it enough context.
Push the conversation
Ask it "push this idea further," "rephrase it in a more direct tone," "give me 3 variations." The 3rd or 4th version almost always beats the first.
If you get stuck
Write to me. I've done these steps hundreds of times, and the block always comes from the same place. I reply to every email.
↗ You just mastered door 1. Ready for door 3?
When you feel like your Chat isn't enough anymore — that you're repeating the same context, that you want Claude to act on your files — you're ripe for door 3. You can go there now, or wait until you've got 2 weeks of Chat use under your belt. But don't let the word "Code" hold you back.
Which door for your job?
No one-size-fits-all answer. Here's what I recommend by profile, with concrete numbers on the gains people see. If you don't recognize yourself exactly, take the closest profile.
Coach / trainer
Trainers equipped with an AI workflow save 10 to 15 hours a week on content creation.
Setup: Chat + Custom Instructions + 1 "My clients" Project. ↗ Level 2 · Claude Code: a CLAUDE.md with your tone of voice + one client folder per person. You ask it to draft the next session, the follow-up email and the recap all at once. Source · Webnyxt 2026Restaurant owner / craftsperson / retail
76% of restaurant owners already use AI. Used well, it cuts the time spent replying to Google reviews by 70%.
Setup: Chat + 1 "Customer reviews" Project. ↗ Level 2 · Claude Code: a "reviews" folder where you drop the week's export. You ask it to sort them, write 30 replies in your tone, and watch for weak signals (a recurring complaint, a product to pull). Sources · Access Group 2025 · Booteek.aiE-commerce seller
DTC brands that use AI for their product pages see an average +18 to 20% conversion rate.
Setup: Chat + one Project per brand or product line + Cowork to process the full catalog. ↗ Level 2 · Claude Code: you point it at your catalog (500 products). It rewrites every product page following your brand book, watches competitor prices, and prepares your acquisition emails — all in one night. Source · EnviveB2B consultant
On writing executive summaries, Claude saves an average of 59% of the time.
Setup: Chat + one Project per engagement + strict adherence to the privacy rules (see next section). ↗ Level 2 · Claude Code: one folder per client with all past deliverables. You ask it for a cross-engagement executive summary, a competitive benchmark, an analysis of your call notes. Sensitive data that stays on your machine. Source · Federal Reserve 2026Comms / creative agency
Agencies that use AI on their creative briefs hit an average ROAS of 3.8× in 6 weeks.
Setup: Chat + one Project per client + Cowork for competitive monitoring. ↗ Level 2 · Claude Code: a client wiki that enriches itself automatically (briefs, transcripts, past deliverables). When the client calls back 6 months later, you have a brief ready in 20 seconds instead of re-reading 40 documents. Source · Ryze AIWhen to move up to Claude Code?
If you recognize one of these 3 signals in your week, you've hit the limit of the Chat and Projects. Your next leap is Claude Code.
You re-paste your context at least 3 times a week
With every new conversation, you re-type your offer, your tone of voice, your strategy. Claude Code reads your CLAUDE.md file at every session: you write your rules once, and they apply for life. Never re-paste again.
Your documents are too big for uploads
You've got a folder of 50 PDFs, 200 notes, a 10,000-row export. Claude.ai hits its ceiling fast. Claude Code reads straight off your disk — no upload limit, you point it at a folder and it processes it.
You want it to act, not just answer
Rename 500 photos, generate 30 PDF invoices, transcribe 4 hours of meetings then send 4 follow-up emails — the Chat answers "here's how." Claude Code does it. It executes, it edits, it sends.
In plain terms
Claude Projects = Claude that answers you. Claude Code = Claude that acts. The day you want the AI to do it for you instead of just telling you how, you climb to step 3.
What you can do without writing a line of code.
Here are 7 things I do, or that an entrepreneur can do, with Claude Code. None of them require knowing how to code. It all gets written in plain English, like talking to a junior coworker who's very fast and never tired.
Tidy up your desktop and your Downloads
"Sort my Downloads folder: group the PDFs into a subfolder, the images into another, and archive anything older than 6 months into an Archive folder." → 200 files sorted in 2 minutes, without clicking through them one by one.
Generate an expense report from a bank export
"Here's the CSV export of my business account. Pull out last month's business expenses, sort them by category (travel, meals, software, etc.), and write me a table ready to send to my accountant." → Replaces 1 to 2 hours of Excel data entry.
Turn a recorded meeting into a recap + follow-up email
"Here's the transcript of yesterday's client meeting. Write a structured recap (decisions, action items, who does what), then prepare a summary email I can send to the client." → Done by the end of your coffee break.
Run an SEO audit of your site, page by page
"Analyze these 10 HTML pages (meta descriptions, titles, internal linking, keyword density). Compare them to the first page of Google for the query 'X.' List the 5 priority improvements." → Replaces a vendor at €1,500 / month.
Analyze 5 competitors in parallel
"Scrape these 5 competitors' sites, compare their offers, their prices, their sales arguments. Put it all in a table, and tell me what sets me apart." → Replaces 3 days of manual research.
Clean up and enrich your CRM
"Here's my CRM export (2,000 contacts). Remove the duplicates, standardize the company names, and for each contact, look up their current job title on LinkedIn. Hand me back a clean file." → Replaces an assistant at €500 / month.
Write a full newsletter from your notes
"Take my notes from the week (these 5 files), cross them with the 10 articles I saved in this folder, write a 600-word newsletter in my usual tone, with a title, a subtitle and three sections." → Replaces a full day of writing.
What these 7 cases have in common
None of them requires writing code. You describe what you want, in plain English, in your own words. Claude Code reads the files on your computer, understands the context, executes. And it asks for your approval before every irreversible action (delete, send, move).
No, I'm not a dev either.
Hi, I'm Jérémy. I'm not a developer. I never went through any tech training. For years, the terminal intimidated me as much as it probably intimidates most of the people reading this page.
And then I decided to get into it. Not because I needed it to code — because I needed it to stop being overwhelmed by AI. I read, I tested, I failed, I started over. Today, Claude Code runs every day on my computer. It handles my watch, it sorts my emails, it writes with me. Without me writing a line of code.
What I learned along the way is that you do not have to be a dev to pull it off. You just need three things: the right information (not the tutorials for pure devs), the willingness to test (even when you only understand 50% of it), and accepting that you'll fail the first 2-3 tries. That's it.
This whole site is built for that. To pass on the right information, in a language that doesn't shut you out, with examples you can apply Monday morning. If I can do it, so can you. There's no reason for you to be left behind.
What I promise you
Every article, every newsletter, every guide I publish starts from the same principle: I tested it myself, as a non-dev. If a tool made me waste 3 hours, I say so. If a method genuinely changes my day-to-day, I share it. No theory, no scam, no "$2,000 course."
The doubts you might have.
Here are the 7 objections I hear the most. I had every one of them too. Here's how I got past them.
"I'm afraid I'll break everything on my computer."
Claude Code asks your permission before every irreversible action (delete, send, edit). You can say no, and it waits. For even more peace of mind, work in a dedicated folder — if something goes wrong, nothing else is touched. And if you're cautious, use Git (a system that automatically saves every version of your files): you can roll back at any time.
"The terminal is for developers."
Not anymore. Claude Code now has a window app (Mac and Windows): zero commands to type, you click, you write like you would to a coworker. Same power, no intimidation. And even in terminal mode, you only learn 3 commands in your whole life: open, close, launch. The rest, you say in plain English.
"I don't understand a thing about code."
No problem — you don't have to understand it. You describe what you want in plain English ("organize my files," "write me a report," "transcribe this video"), and Claude Code translates that into actions. If one day you want to look at what it did under the hood, you can. But it's never required.
"I'm worried about the cost."
The Claude Pro plan is $20 a month and includes Claude Code (no usage billing). For 99% of a non-dev's use cases, you never have to think about the price. You can also set a spending cap in your Anthropic settings to sleep easy.
"It's complicated to install."
Option 1: the Claude Code window app, you download it like any other app (3 clicks). Option 2: the terminal, a single command to paste. And if you get stuck, you can literally ask Claude in the chat to help you install it, step by step. I did it myself the first time.
"It's going to leak my data."
Claude Code runs locally on your computer. It only sends Anthropic what you explicitly ask it to process — just like the Chat. And you stay in control: never point Claude Code at a folder that contains passwords, API keys or sensitive medical data, that's the simple rule. Everything else can go through.
"I don't have a technical project to give it."
You don't need a technical project. You have files, emails, notes, spreadsheets, appointments, repetitive tasks. That's exactly what Claude Code handles best. Look at the 7 use cases above: none is a "technical project," they're all the day-to-day of running a business.
The 5 beginner mistakes.
I made every one of them. You will too. At least, once you know them, you fix them in 10 seconds.
Mistake · Prompt too short
"Write me an email." Result: a generic, bland email that sounds nothing like you. The fix: give the context (to whom? why? in what tone?), the recipient (nervous client? cold prospect?), the goal (a meeting? a follow-up? an apology?). The more you give, the better the answer.
Mistake · No examples
Claude can't guess your taste. The fix: paste 2 examples of what you like (a well-written competitor email, a LinkedIn post that made you react) and one of what you hate. It calibrates instantly.
Mistake · Accepting the first answer
You ask for an idea, you take the first one. It's the most expensive mistake. The fix: ask for 3 variations, keep the best one, send back "push even further in this direction." The 3rd or 4th version almost always beats the 1st.
Mistake · Forgetting the context between conversations
You re-paste your pitch, your offers, your tone of voice with every new conversation. A waste of time. The fix: that's exactly what Custom Instructions (section 3) and Projects are for. You put your context in once, it's reused everywhere.
Mistake · Believing Claude knows everything
Claude doesn't have your business context. It hasn't read your emails, it doesn't know your clients. The fix: you're the one steering, not it. You have to give it the raw material. Simple rule: if a new human on your team couldn't answer without more info, neither can Claude.
What you must never put into Claude.
You may end up working with sensitive data: client information, contracts, strategies, financial data. Before you paste just anything, read this section. It's the most important one on the whole page.
The rules by type of plan
| Plan | Training on your data? | Suitable for client data? |
|---|---|---|
| Claude Free & Pro ($0 / $20 a month) | Yes, since September 2025. 5-year retention. | No for sensitive client data. |
| Claude Team & Enterprise ($20 to $60 per seat) | No by default. Zero-Data-Retention option on Enterprise. | Yes. It's the standard for professionals handling client data. |
| Claude API (pay as you go) | No by default. | Yes, but it requires code. |
The 4 absolute no-gos
The right anonymizing reflex
Before you paste a document, replace the names with "Client A," "Client B," "Prospect 1." You keep the substance and the numbers, you strip out the legal risk. Five minutes of find-and-replace protect you from problems that cost far more.
Legal note · February 2026
A U.S. court confirmed (the US v. Heppner ruling) that conversations with an AI are not covered by professional privilege. If you're a lawyer, a doctor, an accountant or bound by professional secrecy, be especially careful about what you paste, whatever the plan.
And now what?
You've got the basics. Depending on where you are, here's what to do next. I've sorted my articles by level — if you're coming straight from here, start with the "Beginner" column.
If you take away one thing
Don't go looking for the magic tool. Just aim to use Claude one hour a day for two weeks. After 14 days, you'll be more comfortable than 90% of the people around you, and you'll know exactly when to open door 3.
Two paths. Your call.
You can just follow along quietly (my AI watch every Friday), or jump straight into action with the Claude Code guide. Or both. It's for you.
Get my AI watch every Friday
10 AI updates that matter, curated, tested. No ads, no spam. Unsubscribe in 1 click.
The step-by-step Claude Code guide
Install Claude Code in 10 minutes, with screenshots, no scary terminal. And walk away with a first prompt that works.
See the Claude Code guide →