Really useful piece. What stands out is not just Claude Code itself, but the bigger shift behind it. Coding is slowly becoming more accessible to people who can think clearly, give direction, and test well, even if they do not write syntax.
Exactly. Once the model layer becomes easier for everyone to access, the real difference shifts to product judgment. Not just what the tool can do, but what problem it solves, how smoothly it fits into work, and whether people actually keep using it.
Greaaaat article. I've been using Claude Code on a Raspberry Pi for a few months. This article does a much, much better job than I ever could in describing setup, benefits, tweaks, annoyances. Love your stuff!
- you can copy the design code of any website on the internet, feed that code directly into Claude Code, and get an exact copy of the website (with your style based on how you prompt). To do this (Mac) go to a website that you really like, press option + command + i and copy the code under styles.
- install the puppeteer function (tell Claude code you want to install it) to set up an auto screenshot function so you don’t need to manually screenshot errors you find. It will do it automatically.
Good piece — Ruben writes well for a non-technical audience and has clearly put thought into accessibility. A few honest takes:
What’s genuinely useful here:
The CLAUDE.md memory trick is solid and underused by beginners. The “screenshot the bug” workflow is practical. The “describe the outcome, not the steps” prompting advice is correct and worth repeating. These are real habits that save time.
Where it gets a bit sloppy:
He conflates Claude Code (the terminal-based agentic coding tool) with the Claude desktop app’s “Code” tab, which are actually different products. The VS Code + Claude extension workflow he’s describing is Windsurf/Cursor territory more than classic Claude Code. If a reader goes Googling “Claude Code” expecting a VS Code plugin experience, they’ll be confused.
The “100x faster” framing is classic content-creator hyperbole. Skip permissions + auto-accept is genuinely useful, but it’s also where things go sideways for non-technical users who can’t debug a spiral loop or catch a messy codebase. He acknowledges this in the “falls short” section at least — which is the most honest part of the piece.
The honest meta-observation:
This is a newsletter for people who want to feel like they’re building without really building. Which is fine!
There is some bad advice in here, coming from a seasoned engineer. Don’t put all your instructions in a Claude.md or instructions.md file; it will eat up the context window even if not relevant for the task. Use SKILL.md files instead.
You don’t have to be an expert dev to share your experience. But it is a good idea to research best practices before unintentionally giving bad advice on the matter as if you were the SME.
I've been searching for this kind of detailed explanation from a beginner's perspective. Really helpful and starting to learn the basics. Thanks a lot.
I might've missed this in an existing comment but do you have resources you'd recommend for coders/devs who are looking to figure out how best to use tools like claude code (if at all?)? Thanks!
That’s it. No prompt, no setup “Claude.md” instructions. Just /init. It will set up the memory, system, etc. every time you create a new directory or project.
Really useful piece. What stands out is not just Claude Code itself, but the bigger shift behind it. Coding is slowly becoming more accessible to people who can think clearly, give direction, and test well, even if they do not write syntax.
what's left is pure product thinking.
Exactly. Once the model layer becomes easier for everyone to access, the real difference shifts to product judgment. Not just what the tool can do, but what problem it solves, how smoothly it fits into work, and whether people actually keep using it.
Thanks for sharing, love the explanation of Claude Code.
glad you liked it :) do you code?
Hey Ruben. Can you share the my voice , my rules with me? Or a guide to set them up would be great!
sure - check my dm.
Ruben, it is so great, would you share the my voice with me as well?
Thanks for sharing this! super detailed, it honestly made me want to open my mac and sign in right away haha
read every single word. really glad i came across this. also… this made me realize how fast claude is evolving :)
go play with it now :) its honestly super fun.
Greaaaat article. I've been using Claude Code on a Raspberry Pi for a few months. This article does a much, much better job than I ever could in describing setup, benefits, tweaks, annoyances. Love your stuff!
the permissions thing is genuinely annoying :) glad you liked it.
Two other tips:
- you can copy the design code of any website on the internet, feed that code directly into Claude Code, and get an exact copy of the website (with your style based on how you prompt). To do this (Mac) go to a website that you really like, press option + command + i and copy the code under styles.
- install the puppeteer function (tell Claude code you want to install it) to set up an auto screenshot function so you don’t need to manually screenshot errors you find. It will do it automatically.
thanks for the tips :)
I shared with Claude (it’s become habit)
Good piece — Ruben writes well for a non-technical audience and has clearly put thought into accessibility. A few honest takes:
What’s genuinely useful here:
The CLAUDE.md memory trick is solid and underused by beginners. The “screenshot the bug” workflow is practical. The “describe the outcome, not the steps” prompting advice is correct and worth repeating. These are real habits that save time.
Where it gets a bit sloppy:
He conflates Claude Code (the terminal-based agentic coding tool) with the Claude desktop app’s “Code” tab, which are actually different products. The VS Code + Claude extension workflow he’s describing is Windsurf/Cursor territory more than classic Claude Code. If a reader goes Googling “Claude Code” expecting a VS Code plugin experience, they’ll be confused.
The “100x faster” framing is classic content-creator hyperbole. Skip permissions + auto-accept is genuinely useful, but it’s also where things go sideways for non-technical users who can’t debug a spiral loop or catch a messy codebase. He acknowledges this in the “falls short” section at least — which is the most honest part of the piece.
The honest meta-observation:
This is a newsletter for people who want to feel like they’re building without really building. Which is fine!
i explained those in the blog by the way :)
I have found your writing to be inspiring and incredibly useful. Thank you.
really glad to know.
Exactly!!!
There is some bad advice in here, coming from a seasoned engineer. Don’t put all your instructions in a Claude.md or instructions.md file; it will eat up the context window even if not relevant for the task. Use SKILL.md files instead.
Look it up.
I actually saw this yesterday: https://x.com/trq212/status/2033949937936085378.
Reminder: me and my entire community we do not want to code for a living.
I believe you are a dev yourself - so a different audience.
You don’t have to be an expert dev to share your experience. But it is a good idea to research best practices before unintentionally giving bad advice on the matter as if you were the SME.
I've been searching for this kind of detailed explanation from a beginner's perspective. Really helpful and starting to learn the basics. Thanks a lot.
you're welcome, Ed :) happy this guide helped
Very useful thank you.
appreciate it, Robert
Bravo 👍👏
thank you - did you try it?
I might've missed this in an existing comment but do you have resources you'd recommend for coders/devs who are looking to figure out how best to use tools like claude code (if at all?)? Thanks!
hi Jen - i'm not a coder myself, it would be hard to recommend the best resource :)
Grok would be helpful, if you're interested to try it
Thank you for this useful and practical guide. I look forward to diving deeper into Claude Code.
hey - how did it go? :)
The release of Claude Designn just upped the ante
both eats so much tokens ahah
Pro-tip. Start every session by typing /init
That’s it. No prompt, no setup “Claude.md” instructions. Just /init. It will set up the memory, system, etc. every time you create a new directory or project.
gotta love Claude Code commands
Still grappling with claude Cowork, but I've saved this post for later. Thanks, Ruben.
hey, hopefully my Cowork guide could help you out: https://ruben.substack.com/p/claude-cowork-20 :)