You don't need to know how to code

The people who bounce off Claude and the people who build whole businesses on it are not separated by talent. They are separated by about twenty minutes of setup nobody tells you to do.

I’ve watched enough friends open Claude for the first time to know exactly where it goes wrong. They type a question, get a decent answer, think “okay, fancy Google, that’s cool,” and close the tab. They never see the part that actually changes things, because the part that actually changes things is turned off by default until you enable it.

This one’s for you if you have never touched AI, you don’t code, and you want to build something anyway. That was exactly me two years ago. Now I write apps and build custom solutions for small businesses to help them grow and automate to their specific needs. Here is how the whole setup works, step by step, in plain English.

First, what Claude even is

Claude is an AI you text or talk to in normal English. You type what you want, it writes back. No commands to memorize, no syntax. If you can text a coworker, you can use it.

There are three different doors that take you into the same Claude, and beginners get confused about which one to walk through, so here they are:

• The website and apps. claude.ai in your browser, plus a desktop app and a phone app. This is where you start. It’s a chat window. You type, it answers. You can pick up your chats across all three, so a conversation you start on your laptop is right there on your phone later.
• Claude Code. The builder version. It lives in your computer’s terminal (the plain text window that scares people) and it can actually create files, write whole apps, and edit real projects on your local machine. This is the one you graduate to when you want to build, not just chat. It comes with the paid plans, and you do not need it on day one.
• The API. Ignore this for now. It’s the plumbing developers use to put Claude inside other software. Not your problem yet.

Start on the website or app. Everything below happens there.

The actual setup: five steps, most of them under a minute

A fresh Claude account is Claude with some of the best parts switched off. Out of the box it has no memory of you, no home for your work, and no saved way of doing your repeated jobs. Every chat is like starting off with a stranger who has no context on who you are. That’s the “fancy Google” version. Still very powerful and useful on its own, but it’s not an optimal thought partner yet.

Here is exactly how to switch the good parts on, in the order I’d do them. Good news for the person watching their wallet: every one of these is free.

Step 1: Make the account and say something real

Go to claude.ai and sign up with Google or an email address. It’s free, and the free tier is more than enough to start. There are paid tiers (Pro, then Max) for when you’re using it every day and start hitting the limits, but don’t pay for the subscription until it earns it for you.

Now open a new chat, and here is the one thing I’d ask you to do differently than everyone else: do not test it with “hi” or a trivia question. Give it a real task from your life with real context. Something like: “I run a small cleaning business. Write a warm, short email to past clients offering 15% off a spring deep-clean, booking by April 30. Keep it friendly, not corporate.” Read what comes back. It can be a past problem you’ve already solved or a real one you’re facing now.

Claude, and other chatbots, close the gap from Googling “15% off spring cleaning email template” and spending 45 minutes editing, re-writing, and polishing, to being able to ask a direct question and get a direct answer with real deliverables. That gap you just felt, between typing a two-word search and actually briefing it, is the whole skill we’re going to learn here.

Step 2: Tell it who you are (Preferences)

This is the fastest free upgrade there is, and too many users skip it.

Click your initials in the bottom-left corner, open Settings, and find Instructions for Claude. This is a box where you tell Claude, once, who you are and how you like it to work. A few plain sentences is plenty. Mine says roughly: I run a web design and AI business, I write in a direct and warm voice, skip the corporate filler, and never use em dashes, plus a few other quirks that are specific to me. From then on, every new chat already knows what I’ve told Claude.

People also refer to this as “user preferences,” and it’s the manual half of making Claude know you: it’s the stuff you tell it on purpose. It’s on every plan, free included, and it takes two minutes. If you don’t know where to start, like I didn’t, try asking Claude Chat something like:

Help me set up my Claude preferences from scratch. I’ve never done this before, so please walk me through it one question at a time instead of dumping everything at once.

Ask me about things like: what I do for work or what I mostly want help with, how much detail I like in answers (quick and short vs. thorough), whether I want a casual or formal tone, any writing quirks I care about (words or phrases I hate, formatting I like), and anything about my life or interests that would help you give better answers.

After we talk it through, write my preferences up as a clean, organized profile I can paste into my settings. Keep it plain and readable, and explain what each part does so I understand what I’m saving.

Step 3: Turn on Memory

If Preferences is what you tell Claude, Memory is what Claude notices on its own.

In Settings, open Capabilities and toggle Memory on. Now it carries what it learns about how you work from one conversation to the next, without you spelling it out. The projects you keep coming back to, the way you like things done, the details you’d otherwise re-explain every morning.

A thing I assumed wrong for a long time, and want to save you from: memory is free on every plan now, not a paid-only perk. The paid tiers add the ability to search back through your old chats, but the remembering itself costs nothing. And if it ever knows too much or gets something wrong, you can pause or reset it in the same Capabilities screen. I wrote two full episodes on this exact wall, Claude doesn’t know you and then the fix, Claude actually can know you.

You can also “View and manage memory” from your Settings. If there’s anything Claude keeps getting wrong, or not remembering about you that you want it to, you can view and edit the memory directly in Settings or by having a conversation with Claude Chat. Try something like this:

You’ve got some incorrect information in your memory about me that I need fixed. Here’s what’s wrong and what it should say instead:

[List each one, like: “You think I still work at X, but I left in March, I now work at Y.” Or “You have my dog’s name as Max, it’s actually Cooper.”]

Please update your memory so the wrong details are corrected or removed, and confirm back to me exactly what you changed so I know it stuck. If anything I’ve listed is unclear, ask before you change it rather than guessing.

Step 4: Start a Project

A Project is a folder for one thing you’re working on. You drop in the relevant files, notes, and context once, and every chat you start inside that Project already knows all of it.

Think of it like a job site versus a junk drawer. Without Projects, every conversation is a fresh drawer you have to refill, every time. With a Project, you walk onto the same job site every time and all your tools are already there, set up exactly how you like them.

Here’s how: in the left sidebar, click Projects, then New Project. Name it for one real thing. Add a sentence or two of project instructions, and drop in any files that matter (a brand doc, your notes, a spreadsheet). The free plan lets you keep up to five projects, which is plenty to start.

Building a website for your side business? That’s a Project. Planning a garden? Project. Writing a book? Project. If you’ll touch it more than once, it deserves a Project. I went deeper on this in projects versus chats. But for this walkthrough the important thing to know is that you can ask Claude Chat to help you get started (see a theme here?). Try asking something like:

I want to set up a Claude Project for the first time and I need help figuring out how to structure it. Ask me questions to understand what I’m trying to do before we write anything.

Here’s roughly what the project is for: [describe it in a sentence or two, even loosely, like “keeping track of my book club” or “all my work on a client website” or “planning my garden this year”].

Walk me through it a few questions at a time. I want to end up with clear project instructions Claude can follow every time I work in this project, and a sense of what files or documents I should add to give it context. Explain the difference between the instructions and the uploaded files so I understand what each one is doing.

Step 5: Skills and Claude Code, when you’re ready

The first four steps are day one. These last two are for when you’ve been using it a while and feel you’re ready for more.

Skills teach Claude your repeat jobs once. I’ll borrow my own definition from Let’s talk about skills: a skill is a set of instructions you give Claude once that it can pull up and follow whenever the situation calls for it. A saved playbook it runs on demand. Instead of re-explaining how you write a client proposal every single time, you write down how the job is done, once, and Claude reaches for it automatically. I have twenty-five now. Every one started as a thing I was tired of explaining more than once. You don’t build these on day one. You build them the third time you catch yourself typing the same instructions.

In this scenario, I would suggest waiting until a pain point has hit: something in your life or work is taking too long, too many repetitive actions, or too much unnecessary cognitive load. I want to be careful with the framing here, because a skill doesn’t automatically mean there is no manual work involved. It just means the skill is holding the context and the workflow of an existing process to reduce the manual effort, not eliminate it. You can start by asking Claude Chat something like:

I want to build a Claude skill and I need help scoping it before we write anything. Ask me questions to pin down what it should do.

Here’s the rough idea: [describe the task you want the skill to handle, like “run my client build intake” or “draft posts in my brand voice” or “prep me before a sales call” or “meal-plan with two picky toddlers for the week”].

Walk me through it a few questions at a time. I want to nail down three things: when the skill should trigger (the phrasings and situations that should fire it), what it actually does once triggered, and what supporting files or reference material it needs to do the job well. End by drafting the SKILL.md, and be specific about the description field since that’s what controls whether the skill triggers at all.

Claude Code is that builder version we touched on, the one that lives in your terminal and writes real files and apps. When chat stops being enough and you want Claude to actually make the thing you’ve talked about, that’s where you go. It comes with the paid plans, and it’s a whole episode of its own down the line.

To continue my pattern of pushing you toward Claude Chat, you can always start your journey with a conversation:

I use Claude in the regular chat, and I keep hearing about Claude Code but I don’t fully get what it is or whether I’d actually use it. Help me figure that out before I install anything.

Ask me a few questions about what I do and how technical I am, then walk me through: what Claude Code actually is and how it’s different from chatting, the kinds of tasks it’s good for versus tasks the chat handles fine, what I need to have set up before I can try it, and honestly whether it’s worth it for someone like me or if I’d be better off staying in chat for now.

Don’t sell me on it. If it’s overkill for what I do, tell me that.

The mental model that matters more than any button

Here’s the shift, and it’s the whole game: stop simply asking Claude questions about your work and start delegating work to it.

A search engine wants a short query. Claude wants the opposite. It does its best work when you tell it what you’re trying to do, who it’s for, what “good” looks like, and what you’ve already got. The more of that you hand it up front, the less it has to guess. It’s called spec-first, which is a fancy way of saying tell it the whole plan before you ask it to build the thing.

Remember that gap from Step 1, between the two-word search and the briefed task? That’s this. The blank message box invites a one-line question. Resist it. Talk to Claude the way you’d brief a sharp new hire on their first day, not the way you’d Google a phone number. If you’re unsure how to make this shift, ask Claude Chat something like:

I want to get better at delegating real work to you instead of just asking you short questions. I’ve heard the trick is to brief you like a new hire: tell you what I’m trying to do, who it’s for, what a good result looks like, and what I already have, instead of firing off a one-line question. I’m not sure I’m doing that well.

Here’s a task I’d normally hand you as a quick question: [paste the short version, like “write a post about my new project” or “help me plan my week”].

Two things: first, ask me the questions I should have answered up front, the stuff you’d need from me to actually do this well. Then show me the fuller, briefed version of that same request so I can see the difference. Going forward, when I hand you something too thin to do well, tell me what’s missing instead of guessing.

What beginners get wrong

Almost every stall I’ve watched comes down to one of these:

• Treating it like search. One-line questions get generic answers. Oftentimes incorrect ones. Give it context and a goal.
• Never setting Preferences, Memory, or a Project. You’re using the demo version, and you can level up without touching the paywall.
• Waiting to feel ready to build. You learn this by making one real thing badly, not by reading ten more guides. Including mine.
• Trying to build a skill library before you have the friction. Everyone is going to tell you to use skills, and they’re right. But build them for the jobs you actually repeat, once the habit is automatic, not the ones you imagine you’ll repeat someday.

What I’d tell you

Pick one real thing you want to exist. Not a practice project, a real one. The website, the tool, the tracker, the assistant, the thing you keep saying you’ll get to.

Make the free account. Fill in your Preferences. Turn on Memory. Start a Project for that one thing. Brief Claude like a new hire instead of quizzing it like a search bar. Build the actual thing, badly, this week. Right now. (Seriously, stop reading and go do it!) Around the third time you repeat yourself, save a skill.

That’s the entire setup. Everything else on this blog is a deeper cut on one of those moves. You don’t need to know how to code, you don’t have to be an engineer, and you don’t need to know more than this to start. I didn’t know any of it. You need a real thing to build and a free account, and the rest accrues while you work.

Go make the account. Then come tell me what you built.