**Ultimate Claude Code Tutorial: How to Use Claude Code for Beginners in 2026 — Transcript & Summary | SozAI**
Source: https://sozai.app/transcript/claude-code-tutorial-beginners-2026/

A beginner's 2026 guide to Claude Code: installing the desktop app, choosing a paid plan and model, and building a podcast site, logo, and Chrome extension using a plan-then-build workflow.

## Key Takeaways

- Claude Code is accessible via a desktop app with no terminal setup needed, but requires a paid plan.
- Use one focused folder per project and review every diff before approving changes.
- Plan with Opus, build with Sonnet: this balances quality of design against cost and speed.
- Specific, narrow prompts that state current behavior, desired change, and constraints produce safer edits than broad rewrite requests.

## What the video covers

- Claude Code desktop app installs directly now, replacing the older NodeJS/npm/terminal setup, and requires a paid plan (Pro or higher).
- Users should open the Code tab, choose a local, dedicated folder per project, since Claude can read, edit, and run anything inside the workspace.
- The recommended pattern: Opus plans the work (cheap moment to change direction), then Sonnet builds it faster and cheaper per token.
- The first project is a one-page podcast website, with the output previewed in the browser after the build.
- The video promotes AI Master, a platform for generating images, audio, and video, chatting with multiple models, and running an AI-agent channel, with an academy and a 7-day money-back guarantee.
- Visual fixes can be made by dropping a screenshot into the chat and asking Claude to fix the issue, since it sees the image and the project files.
- The second project is a Chrome extension for saving highlighted quotes, loaded via developer mode in Chrome, and built with Sonnet since the scope is small.
- Effective edits state what the code does now, what is wanted, and what must stay unchanged, and users should review diffs before approving.
- Claude can explain existing code in plain English before editing it, helping users understand projects they did not write.
- A third project involves Python packages; the video recommends turning off auto mode for first projects so commands are shown and approved manually.

## Chapters

1. 00:00 Introduction and Installing Claude Code
2. 01:16 Choosing a Folder and Checking the Connection
3. 02:22 Podcast Website: Plan with Opus, Build with Sonnet
4. 03:41 Adding a Logo and Promoting AI Master
5. 04:46 Visual Fixes with Screenshots
6. 05:57 Chrome Extension for Saving Quotes
7. 08:00 Running Commands, Debugging, and Workflow Habits
8. 18:00 Pricing and Next Steps

Answers

## Questions about this video

What plan do you need to use Claude Code according to this video?

Claude Code requires a paid plan, specifically Pro or higher, because the free tier will not open it. The video notes Pro is about $20 a month.

Why does the video recommend using Opus for planning and Sonnet for building?

Opus is a slower, deeper thinker that catches edge cases before files are created, though it costs more per token. Sonnet writes code faster and cheaper, so it suits the building stage after the plan is approved.

How should you give Claude instructions for editing existing code?

The video advises stating what the code does now, what you want instead, and what must stay exactly the same. This keeps Claude from rewriting working parts, and you can check the diff before approving.

## Full Transcript — Download SRT & Markdown

00:00

Speaker A

Claude Code can create files, make changes to your project, and fix errors directly on your computer. In this video, we'll start from scratch, set everything up, build some working apps, and I'll show you the easiest way to start using it. In June, this part took

00:15

Speaker A

three installs, and a terminal window. You installed NodeJS first, then Claude Code through npm, then cursor, so you could actually see your files. You don't need any of that now. Go to claude.com download. Grab the desktop app for Mac

00:29

Speaker A

or Windows. Run the installer and sign in. There's also a browser version of Claude Code. Everything in this video uses the installed app because that's the one that works with folders on your own computer. Before anything else, check your plan. Claude Code needs a

00:44

Speaker A

paid plan, Pro or higher, and the free tier won't open it. Then look next to the send button. That's the model dropown. Since last week, the top option there is Opus 5.5. And right next to it sits Sonnet 5. On Pro, the app now picks

00:59

Speaker A

Opus 5.5 as the default, and you can leave it there for your first questions.

01:04

Speaker A

We'll come back to it in a minute because switching models on purpose is one of the most useful habits in this whole video. Now, the important part, open the code tab, choose local, and select a folder. It can be completely

01:16

Speaker A

empty. From that moment, Claude treats the folder as its workspace, which means it can read, edit, and run anything inside it. So, be picky about which folder you give it. Make a new one for each project instead of pointing it at

01:28

Speaker A

your desktop or downloads. A focused folder gives Claude the right context and keeps it away from files that have nothing to do with the job. To check the connection, type what files are in this project. If it lists the contents or

01:41

Speaker A

tells you the folder is empty, you're set. The first build is a one-page site for a podcast because it's the fastest way to see the whole loop from an empty folder to something in your browser.

01:51

Speaker A

It's also where I want to show you the pattern I use on almost every project.

01:56

Speaker A

One model plans and a different model builds. Opus is already selected in the dropdown. So for the plan in step, you don't change anything. Opus is the slower, deeper thinker. And even after the Opus 5.5 price cut, it costs more

02:10

Speaker A

per token than Sonet. It spends more time reasoning and catches edge cases before a single file exists. For a single page, that's overkill. But you'll use this handover constantly on bigger things. So, it's worth seeing it once on

02:22

Speaker A

something small. I enter the prompt and Opus comes back with a plan and a list of files and nothing has been created yet. Read it. This is the cheapest moment to change your mind because moving a section around in a plan costs

02:35

Speaker A

nothing. Once I'm happy, I switch the drop down from the default opus to sonnet. Then I tell it to follow the plan exactly and build it. Sonnet is the builder. It writes code fast and costs less per token, which makes it the right

02:49

Speaker A

fit for this job. So, Opus designs the blueprint and Sonnet does the construction. To see the result, open the preview or drag index.html into your browser. The page is live and you haven't written a line of it. A page

03:02

Speaker A

like this needs a logo and Claude Code won't draw one for you. So, here's a quick ad for my own platform, AI [music] Master, because this is exactly the gap it fills for us. The show is called The

03:13

Speaker A

Long Conversation. So, I open the image generator and paste that name in. I ask for a minimal logo with two overlapping speech bubbles on a plain background. It has to stay readable, small. I get one option back in about a minute. It works

03:27

Speaker A

fine as a tiny icon. So, I drop the file straight into the project. The image tool is only one window of it. In the same workspace, you get a chat with the top models, Claude, Chat, GPT, Grock, and Gemini, and you switch between them

03:41

Speaker A

without juggling four subscriptions. [music] It works out cheaper per token than paying for each one separately.

03:47

Speaker A

Image, audio, and video generation live in the same place, and they're cheaper there, too. You can build a consistent character that looks the same across every generation. Then you can publish that character on the platform and let it earn for you and share anything you

04:01

Speaker A

generate so it works for your reach. The biggest piece is viral. It's where you and your team run a full channel with AI agents. They write the scripts and descriptions, edit the videos and images, and make the thumbnails. On top

04:14

Speaker A

of that, there's the academy with more than 200 lessons and about 30 hours of material. More than 13,000 people already use AI Master, and the results page is full of their cases with screenshots straight from the site. The

04:27

Speaker A

path is simple. You open aimaster.me, pick the annual subscription, which usually comes with a discount, and you're inside the same day. If it's not for you, there's a 7-day money back guarantee. The link is in the description. Back on the podcast page,

04:42

Speaker A

let's say the listen button is the wrong color or the episode titles feel heavy.

04:46

Speaker A

You don't need to know which file controls that. Take a screenshot of the part that bothers you. Drop it into the chat and write fix this. Claude sees the image and has the whole project on disk at the same time so it can find the

04:59

Speaker A

right file on its own and edit it. You also don't have to explain which button you mean. When it fixes one element, it will often update the matching ones, too. So the whole page stays consistent instead of half old and half new. It's

05:12

Speaker A

the fastest way I know to go from something looks off to fixed. The second project lives inside your browser. You highlight a sentence on any page, rightclick, and save it as a quote. The extension keeps the text, the page

05:25

Speaker A

title, and the link and shows everything in a small pop-up that's handy for research scripts, and anything you read with a pen in hand. Make a new folder and point the app at it. This time I skip the planning step and switch the

05:39

Speaker A

drop down straight to Sonnet because the scope is small and clear. A Chrome extension is only a handful of files.

05:45

Speaker A

Knowing when not to use the expensive model is half of the pattern. Claude writes the manifest, the background script, and the pop-up, and you review them in the diff like before. Installing it takes about [music] a minute. Open

05:57

Speaker A

Chrome, go to extensions, turn on developer mode in the top right, and click load unpacked. Then select your project folder. Now open a long article, highlight a line that matters and hit save quote. Do the same on two or three

06:11

Speaker A

other pages. Then click the icon. Every quote sits there with its source link and one button copies the whole list as clean notes you can paste anywhere.

06:20

Speaker A

[music] After a week of reading, that single list in the popup gets long and messy. Before you touch anything, ask Claude to explain the part you care about. Something like, "Explain what happens after I click save quote." in

06:33

Speaker A

plain English. It reads the code and walks you through it without making you read every line. That's how you get familiar with a project, especially one you didn't write or wrote weeks ago and forgot. Then make the edit narrow. A

06:46

Speaker A

request like rewrite this gives Claude permission to touch everything, including parts that already work. A better edit says three things. It says what the code does now, what you want instead, and what must stay exactly as it is. For example, the pop-up shows all

07:01

Speaker A

quotes in one list. I want them grouped by website and nothing else changes, especially the context menu and the markdown button. Claude changes the one piece and leaves the rest alone. And the diff shows you that before you approve.

07:14

Speaker A

On bigger projects, this habit is the difference between a small update and an accidental rewrite. Step back for a second and look at what you have. A working Chrome extension sits in your browser and you didn't write a line of

07:26

Speaker A

it. That's the real shift. Claude Code lets you solve the one annoying problem that no app on the market solves for you. Say you run a small online store and every Monday you copy order numbers from confirmation emails into a

07:40

Speaker A

spreadsheet by hand. No off-the-shelf tool fits that exact routine. A small script or extension that does it for you is an afternoon of prompting. And whatever task keeps slowing you down works the same way. The third project doesn't have a screen at all. It's a

07:54

Speaker A

Python script that watches your downloads folder. Every time a new file lands there, it checks the file type and moves it into the right subfolder. PDFs, images installers spreadsheets everything else. If two files share the same name, it adds a number instead of

08:10

Speaker A

overwriting one of them. It all runs locally and nothing gets uploaded anywhere. This one touches more moving parts. So, I go back to Opus for the plan. I'm inserting a prompt like this.

08:20

Speaker A

Opus lays out the dependencies and the steps. Then I switch to Sonnet and say build it. This is the moment Claude stops acting like a chat. It doesn't hand you commands to paste into a terminal because it runs them itself. It

08:33

Speaker A

wants to install the Python packages the plan called for. I switched auto mode off for this first project. So it shows you the exact command and waits. You read it, approve it, and watch the output scroll by in the same window.

08:45

Speaker A

Then it starts the watcher and again it asks first. That approval step is where you stay in charge. So, it's worth having a rule for it. Installing a well-known package, starting a local server, or running tests is routine, and

08:59

Speaker A

you can approve those quickly. Anything that deletes files, overwrites data, or pushes something live deserves a slower read. If you can undo it easily, the risk is small. If you can't, take 10 seconds and check what's about to run.

09:12

Speaker A

And if something looks wrong halfway through, hit the stop button. It interrupts the command right there. So, you still decide what runs on your machine and when. When the build finishes, Claude starts the watcher and tells you it's running. Drop a random

09:25

Speaker A

mix of files into the folder, a PDF, a screenshot, an old installer. A second later, each one lands in its matching subfolder, and a file with a name that's already there keeps its own number instead of replacing the original. At

09:39

Speaker A

some point, a script like this will break. Usually, you get an error message, and the rule for that is simple. Paste the whole thing. Write something like, "I get this error when a file lands in the folder. Here's the

09:50

Speaker A

full message." Claude reads it, traces it back to the right line, and proposes a fix you review in the diff. Notice what your job has turned into over these three projects. You aren't writing code.

10:00

Speaker A

You decide what gets built, what gets cut, and whether the result is actually any good. Claude builds fast, but it has no opinion on whether the file names make sense or the podcast page looks cheap. That judgment stays with you.

10:13

Speaker A

People without a coding background who get the most out of this tool stop worrying about syntax. They put that energy into knowing what good looks like and the last project is a nice small place to practice it. The last build is

10:26

Speaker A

small and personal. It's a subscription tracker that runs entirely in your browser. You add every service you pay for with the price and the renewal date and the app shows your monthly total and what renews next. Everything stays on

10:40

Speaker A

your computer with no account and no server behind it. The scope is tiny, so I switch to sonnet and describe it in one go. I'm also adding one constraint on purpose. Everything goes into a single HTML file because that keeps the

10:53

Speaker A

project from growing into something it doesn't need to be. When it's done, you doubleclick the file and the tracker opens in your browser. You don't start a server and you don't run a build step.

11:03

Speaker A

It's one file you can email to yourself or keep in a cloud folder. The data stays in that browser on that computer.

11:10

Speaker A

So, if you open the file somewhere else, you start with an empty list. That's a fair trade for something this simple.

11:16

Speaker A

And if it ever bothers you, adding an export button is one more prompt. Here's where the fun starts. The tracker works, but I want a small upgrade. I asked Claude to paint every subscription that renews in the next 7 days red and to

11:30

Speaker A

show the yearly total next to the monthly one. That second number is usually the one that hurts. Claude adds both and the diff shows it touched the list styling and the total calculation and nothing else. That's the loop you

11:43

Speaker A

fall into with this tool. You ship a working version fast, then keep refining it one short request at a time until it feels like yours. Each request is small so each diff stays easy to read. Once you've built one tool like this, a lot

11:56

Speaker A

of small ideas become a weekend project. A reading list can keep the books you want, the ones you finished, and a oneline note on each. A packing checklist can remember what you always forget and build a fresh list for every

12:09

Speaker A

trip. A freelance hours log lets you start a timer per client and shows at the end of the month what to invoice. A plant watering page can simply tell you which pot is thirsty today. Pick the one you'd open every week. That's your next

12:22

Speaker A

folder. By the fourth project, you'll notice a pattern in your own prompts. You keep telling Claude the same things.

12:28

Speaker A

Use plain JavaScript. Don't add libraries. Name things clearly. Claude doesn't carry memory from one session to the next on its own. So every new chat starts from zero. The fix is a file called claude.mmd that sits in the root

12:41

Speaker A

of your project folder. Claude reads it whenever it works on that project. So your standing rules are already in context before you type the first request. You don't have to write it from scratch. Type in it in the chat and

12:54

Speaker A

Claude looks through the project and drafts a claude.md based on what it finds. Then you open the file and add your own rules. For the subscription tracker, I'd keep it to four lines.

13:04

Speaker A

Plain JavaScript and no frameworks, full words and function names, no abbreviations. Everything stays in one file. Never add a library without asking me first. Now ask for a new feature like a category tag for each subscription and Claude follows those rules without you

13:21

Speaker A

repeating a single one. Even three or four short rules make a visible difference. So do this near the start of every project. So far we've worked in small steps. You ask, you review the diff, you approve, you ask again. That's

13:35

Speaker A

the right rhythm while you learn. But on anything that touches more than two or three files, Claude can start writing and realize halfway through that it needs to restructure what it just built.

13:45

Speaker A

Plan mode prevents that. When it's on, Claude reads the project and writes a plan without changing any files. You read the plan, fix what's wrong, and only then let it build. It's the same idea as the opus planning step built

13:59

Speaker A

right into the tool. The next step up is to give Claude a whole multi-step job and let it work through it without waiting for you after every file. What makes that work is the brief. Make this better leaves everything open. A good

14:12

Speaker A

handoff has four parts. It names the goal, the constraints, what counts as finished, and what you specifically don't want added. Here's the kind of brief I mean for a small content calendar built in a fresh folder. After you send it, you stop steering. Claude

14:27

Speaker A

plans the app, creates the structure, then builds the ad form, the week view, the publish toggle, and the delete button in order. This is where auto mode earns its place because routine commands go through without you, and only the

14:40

Speaker A

risky ones stop. When it reports that it's finished, your job switches to review. Open the preview and test the four actions from the brief. If one of them misbehaves, that becomes your next prompt. The whole first version came

14:52

Speaker A

from one message because you told Claude what done looks like. There's one more piece that arrived this month, and it's still in beta. New projects lets you run several of these handoffs at the same time in the cloud, each in its own

15:04

Speaker A

thread. So, the content calendar builds in one thread, while another one writes a small export feature for the subscription tracker. I wouldn't start there as a beginner. Every parallel job is one more result you have to review and review is still your part of the

15:19

Speaker A

deal. Once a single handoff feels routine though, it's the obvious next step. I want to be straight about the limits. Clawed code will not replace an engineering team on a banking backend or a medical platform with strict security

15:32

Speaker A

rules. That kind of work needs deep architecture decisions and years of domain knowledge. And the desktop app didn't change that. What it did change is who gets to start. If there's a specific problem in your workflow that no app solves, this is where you begin.

15:47

Speaker A

And if you have an idea and you're not sure it's worth paying a developer for, build a rough version in an afternoon, show it to a few real users and decide from there. If you only need a clickable

15:58

Speaker A

mockup to show someone, the regular Claude chat has its own artifacts and it builds one without any project folder.

16:05

Speaker A

Most of the classic mistakes we already covered on the way. vague prompts, skipping claw.md and asking for a rewrite when you need one change all have their fix in this video. Two traps are left and both come from long

16:18

Speaker A

sessions. The first one is letting a conversation grow forever. After a few hours, the context gets crowded and Claude starts repeating itself or forgetting decisions it made earlier.

16:29

Speaker A

Type compact in the middle of a session and Claude summarizes the important decisions while dropping the noise. The second trap is fighting the same bug three times in a row. If two attempts to fix something didn't stick, the

16:41

Speaker A

conversation itself is probably working against you because it's full of wrong guesses. Start a fresh session instead.

16:48

Speaker A

Describe the problem cleanly. Paste the full error or describe the exact symptom and point Claude at the file where it happens. A short, clean context almost always beats a long one full of failed attempts. Your claude.md comes with you

17:02

Speaker A

into the new session, so you lose nothing that matters. Most tutorials skip this part. Claude code comes with the paid Claude plans and the entry point is pro at about $20 a month. That gives you a predictable bill and for a

17:15

Speaker A

few sessions a week, it's the right place to start. If you keep hitting the usage limits, the max plans give you much more room starting at around $100 a month. Most beginners won't need that for the first few months. There's also

17:27

Speaker A

the API route where you pay per token. It sounds cheaper, but a long planning session on Opus can cost a few dollars on its own. So, switch to the API only once you know your real usage. The desktop app is the update that matters.

17:41

Speaker A

The setup that used to lose half the audience in the first 5 minutes is gone.

17:45

Speaker A

The one default I change right away is auto mode because a beginner should see every command before letting Claude skip the questions. The Opus plans sonnet builds pattern still holds and with Opus 5.5, the planning step got cheaper. If

17:59

Speaker A

you want the same kind of workflow for images, video, and your channel, AI Master is linked in the description.

18:06

Speaker A

Now, make an empty folder and build something small.

Topics: Claude Code beginner tutorial AI coding Opus Sonnet Chrome extension vibe coding AI Master Anthropic Claude 2026

Study this video

- [Flashcards from this video](https://sozai.app/tools/flashcard-maker/?from=transcript&id=94918)
- [Quiz on this video](https://sozai.app/tools/quiz-generator/?from=transcript&id=94918)
- [Crossword from this video](https://sozai.app/tools/crossword-maker/?from=transcript&id=94918)
- [Word search from this video](https://sozai.app/tools/word-search-maker/?from=transcript&id=94918)
- [Study guide from this video](https://sozai.app/tools/study-guide-maker/?from=transcript&id=94918)

Free, made by AI from this transcript. No signup.


---
This is the markdown twin of https://sozai.app/transcript/claude-code-tutorial-beginners-2026/ — the same content, without the markup.
Published by SozAI (https://sozai.app). Reuse and quotation are allowed with attribution and a link back.
Machine-readable index: https://sozai.app/llms.txt · data API: https://sozai.app/api/
