I typed one sentence into my terminal. Thirty-four seconds later, a working program opened in my browser — a polished HTML report analyzing every file on my computer. I didn’t write a single line of code. I didn’t copy anything. I didn’t even know what Python was doing under the hood. That’s what a Claude coding agent does, and it’s completely different from chatting with Claude in a browser.
I’m not a developer. I’m an SAP consultant who’s been testing AI tools for a blog series — running the same prompts through ChatGPT, Claude, Gemini, and every coding assistant I could find. But all of that was browser-based: you ask a question, the AI shows you code, and it’s up to you to figure out what to do with it. Claude Code is the first time I’ve used an AI that does the work itself.
First-time experience documented in March 2026 using Claude Code with a Claude Max subscription. I have zero Python knowledge and no development background — I’m an SAP consultant. Every step in this article, from installation to running the program, was done with browser Claude’s help in a separate window. All times mentioned include the time I spent asking browser Claude for explanations. Setup and testing took approximately 30 minutes total. AI was used only to lightly copyedit this article’s prose.
What is a Claude coding agent? Browser Claude vs Claude Code
If you’ve used Claude in a browser — which I’ve done extensively across eight comparison posts — you know the workflow: you type a prompt, Claude shows you code in a chat window, and then you have to copy it, create a file, paste it in, save it, open a terminal, and run it yourself.
Claude Code skips all of that. It’s available across the terminal, VS Code, a desktop app, and even the browser — but the core idea is the same: it directly creates files, writes code, and executes programs on your computer. For this first test, I used the terminal CLI, which is why my experience felt so different from chatting with Claude in a browser window.
Here’s the simplest way to think about it:
| Browser Claude | Claude Code | |
|---|---|---|
| You say | “Write me a script” | “Write me a script” |
| It does | Shows code in chat | Creates the file, writes the code, runs it |
| You do next | Copy, paste, save, figure out how to run it | Nothing — it’s already done |
| Result | Code on your screen | A working program on your computer |
That difference sounds small. It’s not. For someone who doesn’t write code every day, the gap between “here’s some code” and “here’s a working program” is the gap between useful and useless.
For the technical context on how Claude Code compares to other coding agents like Cursor and Devin, see my Best AI Coding Agents guide.
Setting up Claude Code — what actually happened
I’ll be honest: the setup wasn’t seamless for me. I spent 10 minutes getting Claude Code installed, and about half of that was because the installation path I followed required Node.js — something I’d never heard of. (Anthropic now also offers native installers that may simplify this step.)
Here’s what happened, step by step.
I opened my terminal — which already felt unfamiliar. I’m used to graphical interfaces, not command lines. The installation command itself was simple enough, but I didn’t understand what it was doing. So I did something that felt ironic: I opened browser Claude in another window and asked it to explain each step.
“What does this command do?” “What is Node.js and why do I need it?” “Is it safe to run this?”
Browser Claude walked me through the Node.js installation, explained what npm means, and reassured me that the install command was from Anthropic’s official source. Five minutes on Node.js, five minutes on Claude Code itself, and I was in.
The first screen after installation was… underwhelming. A text prompt in a terminal window. No buttons, no menus, no visual cues. I genuinely didn’t know what to do next.
Claude Code’s first screen — a terminal prompt and nothing else. For a non-developer, this is intimidating.
What you need before starting
- A paid Claude subscription — for individual users, Claude Pro ($20/month) is the lowest tier that includes Claude Code access. There’s no free tier
- Node.js may be needed depending on your installation path (I needed it; native installers may not require it)
- A terminal — Terminal on Mac, PowerShell or WSL on Windows
- Patience — if you’re not a developer, budget 10 minutes for setup. The official quickstart guide covers the basics
“Make me a program” — my first Claude coding agent prompt
With Claude Code running, I typed this prompt:
Create a simple Python script that analyzes a folder and generates
an HTML report showing: total number of files, file types breakdown,
and the 5 largest files. The report should open automatically in
a browser.
The actual prompt I typed — this is everything I did. One sentence, and Claude Code took it from here.
What happened next genuinely surprised me.
Step 1: Claude Code asked for permission
Before doing anything, Claude Code asked: “Can I create a file called folder_analyzer.py?” This was my first permission request, and I had no idea what it meant. Was it going to modify my system? Could it break something? I switched to browser Claude again and asked what this permission meant. Once I understood it was just creating a new Python file in my current folder — not touching anything else — I approved it.
Claude Code’s permission request — it asks before creating or modifying any file on your computer.
Step 2: It wrote the code
In 34 seconds, Claude Code wrote a 108-line Python script. I watched the code appear line by line in my terminal. I didn’t understand most of it — I don’t know Python — but the speed was impressive. It wasn’t just pasting a template; it was building something specific to my request.
Claude Code writing 108 lines of Python in 34 seconds — building the script in real time.
Step 3: It asked to run the script
Second permission request: “Can I execute python folder_analyzer.py?” Again, slightly scary — it wanted to run code on my computer. But I’d already confirmed with browser Claude that this was safe, so I approved.
Step 4: A program opened in my browser
And then it happened. My default browser opened, and there was a fully formatted HTML report showing every file in my folder — total count, file type breakdown, the five largest files, all laid out in a clean design.
I didn’t create a file. I didn’t open a code editor. I didn’t Google “how to run a Python script.” I typed one sentence, approved two permission requests, and got a working program.
The HTML report Claude Code generated — a working program created from a single chat prompt.
That moment — seeing a real program I never could have built myself appear in my browser — was when I understood what “coding agent” actually means. It’s not a better chatbot. It’s an AI that builds things.
“Add dark mode” — modifying a program in 5 seconds
This is where the claude coding agent surprised me most. The report looked good, but I wanted to push it further. I typed one more sentence:
Five seconds of work — one line typed, and Claude Code handles the rest.
Three minutes later — of which my total contribution was five seconds of typing and one permission approval — the HTML report now had a toggle button in the top corner. Click it, and the entire report switched to a dark color scheme. Click again, back to light. Fully functional.
This is the part that really changed my perspective. Building something from scratch is impressive. But being able to iterate on it with a single sentence — “make it dark mode,” “add a chart,” “change the layout” — means Claude Code isn’t just a one-shot tool. It’s a conversation with a builder. You don’t start over. You keep talking, and the program keeps evolving.
The dark mode toggle Claude Code added in 3 minutes — my contribution was one sentence and one permission click.
The same prompt in browser Claude — spot the difference
To make the comparison concrete, I opened browser Claude and typed the exact same prompt.
Claude responded with a Python script in a code block. Clean code, probably good quality. But I stared at it for a moment, genuinely unsure what I was supposed to do next. Copy it where? Save it as what? Run it how? I’ve been using Claude for months — for writing, research, comparisons. This was the first time it made me feel like the problem was me. The explanation assumed I knew how to create a .py file, save it, and run it from a terminal. Since I’d never told Claude I don’t know Python, it didn’t explain those steps in beginner-friendly detail.
The bottom line: with browser Claude’s response, I could not have created this program. Not without significant additional help. With Claude Code, I had a working program in under a minute.
The same prompt in browser Claude — code is displayed in chat, but as a non-developer, I can’t do anything with this.
The real difference, summarized
| Claude Code | Browser Claude | |
|---|---|---|
| Time to working program | ~1 minute | Impossible without coding knowledge |
| Files created | 1 (automatically) | 0 (you have to create them) |
| Code executed | Yes (with permission) | No (you run it yourself) |
| Result visible | HTML report in browser | Code in a chat window |
| Coding knowledge required | None | Yes — need to save, navigate, and run |
| My honest reaction | “Wait, it actually works?” | “What do I do with this?” |
Is the Claude coding agent worth $20/month?
Let me start with the part nobody talks about: the fear. Every time Claude Code asked for permission to create a file or run a command, I felt a flash of “what if this breaks something?” For developers, that’s probably normal. For me, it felt like handing my car keys to a stranger. The permission system helped — Claude Code asks before every action — but the anxiety was real. And it’s worth acknowledging, because it affects whether you’ll actually use this tool or just install it and forget.
With that out of the way — I’m on the Claude Max plan ($200/month), which is overkill for what I just did. For individual users, the minimum is Claude Pro at $20/month, which includes Claude Code access.
Here’s my honest take after 30 minutes:
Worth it if you regularly need small utilities, scripts, or tools and you’re not a developer. The folder analyzer I built took one prompt. Doing the same thing manually — even with browser Claude’s help — would have taken me an hour of Googling, copying, troubleshooting, and frustration.
Not worth it if you only use Claude for chatting, writing, or research. Browser Claude handles all of that for free. You’re paying for the “agent” part — the ability to create and run programs on your computer.
The irony I noticed: I used browser Claude to understand what Claude Code was doing. The two tools complement each other perfectly — browser Claude for learning and understanding, Claude Code for building and executing. In my previous testing, browser Claude won 2 out of 3 Python coding tests for code quality. Claude Code takes that same quality and removes the “now figure out how to use it” step.
Frequently asked questions
Do I need to be a developer to use Claude Code? No — and I’m proof. I don’t know Python, I’d never used a terminal for coding, and I needed browser Claude to explain the setup process. Despite all that, I had a working program in under a minute after installation. The permission system means Claude Code can’t do anything without your approval, which helps with the trust issue. That said, if something goes wrong, debugging without coding knowledge is hard. I’d recommend having browser Claude open in another window as your “translator.”
What’s the difference between Claude Code and Cursor? Claude Code runs in your terminal and focuses on giving the AI maximum autonomy — it plans, writes, and executes with minimal input from you. Cursor is an IDE (code editor) with AI built in, designed for developers who want AI assistance while they write code themselves. For non-developers like me, Claude Code is more accessible because you don’t need to understand the code editor interface. For a full comparison, see my Best AI Coding Agents guide.
The bottom line
Thirty minutes. That’s all it took to go from “what is a claude coding agent?” to “I just built a program with a dark mode toggle.” Ten minutes for setup (half of that was Node.js). One minute for the initial build. Three minutes for the dark mode upgrade. And the rest was me exploring, comparing with browser Claude, and picking my jaw up off the floor.
The Claude coding agent isn’t a better chatbot. It’s a different category of tool. Browser Claude tells you how to build things. Claude Code builds them. For a non-developer, that’s not a small difference — it’s the difference between “I could theoretically do this” and “I just did this.”
Will I use it every day? Probably not. Will I use it every time I need a small tool, a data report, or a quick automation? Absolutely. And I’ll keep browser Claude open in the next tab to explain what’s happening — because understanding what the AI is doing still matters, even when you don’t have to do it yourself.
This was my first 30 minutes with Claude Code. I’m planning to explore more — building a real project, testing its limits, and finding out where it breaks down. Stay tuned.
Since writing this, I’ve built something more practical — see my Claude Code Tutorial where I created a blog SEO checker. For the current recommended setup method, see my Install Claude Code guide. And for tools with a lower barrier to entry, check out my Claude Code Alternatives comparison.