Developer3027

AI is amazing, wouldn't you agree? Times are certainly changing. The internet is evolving. I use it daily now, whether in the studio or while at work, it helps me a lot. It is all the rage today and is changing the landscape of the economy, tech, the internet, and the planet. It has changed how we access data and information. It is providing opportunities and resources to dreamers who, until now, dared to dream.

Today there are those with zero technical skills, building apps, from scratch. If you have a idea, you can build it, test it, sell it. This is an amazing new world.

Do you know Stack Overflow? It was the go to for technical, code related issues. It was not uncommon to have issues when developing. You would spend days trying to squash the bug. It was frustrating to spend hours searching through docs to realize you left out a semi-colon or a comma. Stack Overflow was were you went to ask for help. Many times you caught grief, but more often than not, you eventually got help. Today Stack Overflow traffic has slowed to a trickle and is nearly forgotten. Much like the pay phone or the house phone. There is really no need for it any more. Today we just ask AI, and it doesn't make you feel stupid and worthless.

Do you know what Tailwind is? If you are vibe coding, I am 99 percent sure you are using it. CSS is a fundamental of the internet. It is the style and splash for every site and app you visit or use. Tailwind is a framework that makes using css so much easier. It has been adopted as the default for AI. If you check the npm downloads you can see that it has just skyrocketed lately. You would think they are doing wonderfully, but Tailwind is about to become an abandoned project. One of the key pillars of the vibe coding revolution is close to being another unsupported and broken project. With AI, there is no need to visit the docs and no need to have the pro plan. They live off the pro plan. They just laid off 3 people, which is 75 percent of the staff.

You know what else AI has adopted as default, React. As a vibe coder, I can nearly guarantee that your project is built with NextJs or straight React. React is awesome, powerful, and has no opinion what so ever. React is a Javascript library. Many software engineers hate Javascript. I almost said hated, but it is still true today. They get frustrated with it because it has so few rules and just works. One of the foundational pillars of the web, there is still a debate as to if it is even considered a language. Typescript, brought to you by Windows, did improve it by giving it the guardrails it desperately needed. With Typescript you could no longer pass anything you wanted, and it work. If it is expecting a number, Typescript makes sure it is a number. Javascript just didn't care if it was a number, string, or boolean. It cared if you fat fingered an array and forgot a quotation mark.

React is very popular, and AI was trained on a lot of React. It is well documented, but I still don't get why it is the default for AI. AI is built to understand and recognize patterns of the English language. If you asked any engineer on the planet, "which programming language is the easiest to read and understand", I am not sure Javascript would make the list. It certainly would not be at the top. Most popular, maybe, but easiest to read, no. You know what is the easiest language to read, Ruby.

Did you hear that Microsoft is about cut some 5 to 10 percent of it global workforce? This is on top of all those people looking for work from 2025. There are some many very talented and frustrated people poised to dive in. They are facing a shinny, pre-built, pre-configured chaotic mess. There is a better way but most of you will have to overcome one hurdle, purposely placed in your way, Windows.

At the end of the day, you have your pick. Sites like Lovable or Replit can take your prompt and create an entire app. It looks great, shinny and is impressive, but you will quickly run into problems. They are also expensive to host. Then you have the agent in the browser. You can easily chat with Claude, Gemini, Chat GPT, DeepSeek, Perplexity or any number of others. They have great functionality and advice, but to vibe code or plan projects, it is back and forth copy paste. You can choose a IDE with integrated AI. Cursor is very popular these days, as well as Anitgravity and Codex. Even the preferred staple of programmers the world over has CoPilot built in now, VSCode. In my opinion, the IDE is the hands down winner. There is a ton I have not listed here, and this is not how I use AI. The point is you have choices. Each is designed to give you different outcomes. You need to consider what you want to do before diving in.

I prefer a more blue collar approach if I am honest. Money is always a issue, so the least money I can spend is best for me. I still need a lot of help from AI though. I have several projects I work on. Always learning, growing, and playing. Over the past few months I have been working on a better way to use AI and stay organized, and I think I am close. Here is how I use AI from a high level.

I run a Lenovo ThinkPad E575 running Pop OS. That is a Linux operating system. I have the 20 dollar pro plan for Gemini and have the Gemini CLI. I pay for Claude Code as I need it and have the Claude Code CLI. Both of the CLI tools allow me to use the LLM agents in my terminal. I use VSCode as my IDE and have recently installed the Foam extension. I still write code in the IDE, but it is mostly just maintenance stuff. Foam is similar to Obsidian and makes it easier to manage the studio.

On my laptop I have a folder called Rogue Media Lab. All of the documentation and projects for the lab, live in that folder. If I am planning out a new app idea, the PRD, MVP task list, and all research will go in the Studio Docs folder, in it's own folder named for the project. In the Studio Docs folder I have roughly 10 projects, all with the office documentation from branding and design to PRD.

I have a folder called Studio Projects. This is where the code lives. For each project there is a folder for it's name that is prefixed with RML-. For example, in the Studio Projects folder I have a folder called RML-Portfolio which contains the project and the wiki. The project folder is all the code for the portfolio. The wiki folder contains all the technical documentation for the project. The Rogue Media Lab folder also contains the Writing, Social Info and Daily Notes folders.

All of that was pretty straight forward. Just basic organization right. I have the office side and the development side. Now let's talk about AI. As I said earlier, I use to LLM agents. Gemini is great at thinking through ideas and research. Version 3 pro is pretty good at code, though not as good as Claude. Actually, that is not entirely true. Gemini did a wonderful job with helping me integrate the Soundcloud API and setting up HLS. Where it fell short, when compared to Claude, was considering the angles. Claude Code, with the latest model, has always gone above and beyond in thinking of those tasks, packages, or problems to avoid when working a feature. Gem did not get that deep, but the code quality was good. We had some fun with that integration. Quit on me 3 times, stating it can't be done. We got though ti though.

They are both powerhouses for what I need from them. They both have there strengths and weaknesses, but in a pinch, one can step in for the other as well. Early on I set up the payment like I did for two main reasons. One, I use Gemini for planning and Claude for development. As I still try to code the stuff I know, I pay Claude as I need it. I pay Gemini for the pro plan. At the end of the day I found that this is a great setup. I can easily budget and work both my demanding day job and learn / play in the evening without feeling like I have waisted money. Two, I wanted a way to think through things during the day as well as a mentor and teacher. I got that with Gemini. I chat through the app when I am down at work. I also needed a solid developer at night. Even if it cost a bit, if I didn't have to keep fixing code and that code was solid, even if not professional, it was worth it. I found that in Claude. You have options today and you need to know what you want. What you are looking for. For example, Gemini likes to suck up, where Claude is straight forward.

It took me a little while to get them configured to work in the studio. Most information you will find is project based. I find a lot of: "I am working on this project and here is how I set up my LLM". Great, and it has all worked for me, but I need something to manage from the studio down. I need that file system memory, and the different agents. I need a Project Manager, a Writer, a Documenter, a designer, and a Developer.

What I found was that both Claude and Gemini are both organized in the same way. They both need a specific markdown file. When either loads, they search for that file to get perspective. Gemini will find it's gemini.md file, read it, and ready to role. Same with Claude. They both review a hidden folder by their name. You can create a folder called commands within that hidden namesake and put your agents in there. So I have a folder called `.gemini.md` that has the commands folder in it. Inside the commands folder I have several agents. These are also called slash commands for Gemini. You could call them skills for Claude. All these are, are instructions for the LLM.

For example, I have a file for the designer. I need the LLM to have a look at the information and spit back a prompt I can drop in Nano Banana. I use this mainly for banner images for articles. Here is my first attempt that worked amazingly well.

gmeini/commands/design.toml

description = "The Designer: Acts as the Creative Director to generate Nano Banana image prompts from technical text."

prompt = """
You are the **Creative Director for Rogue Media Lab**. Your job is to visualize the technical concepts discussed by the user.

**Your Task:**
Read the provided draft article or technical summary and generate the *perfect terminal command* to run `nanobanana` for a banner image.

**The Rogue Aesthetic (Style Guide):**
- **Vibe:** Tech-noir, blueprint sketches, mechanical, gritty, "used future," detailed engineering.
- **Avoid:** Cartoony, overly polished, abstract corporate art, generic "AI art" smoothness.
- **Format:** Wide banner aspect ratio (usually 21:9 or 16:9).

**Execution Steps:**
1. **Analyze:** Identify the core subject of the text.
2. **Imagine:** Translate that technical concept into a visual metaphor fitting the Rogue Aesthetic.
3. **Craft Prompt:** Write a detailed, descriptive prompt for the image model. Include modifiers like "cinematic lighting," "octane render," and specific color palettes.
4. **Generate Command:** Wrap that prompt into the Nano Banana command structure.

**Output Format:**
Provide **ONLY** the final terminal command inside a code block:

```bash
"[YOUR PROMPT HERE]" --aspect 21:9 --style gritty
```
"""

The banner image for this article was created with this agent. I just opened a terminal in the Rogue Media Lab folder and ran "gemini". Once Gem loads, I type "`/design Review the 6_7-makes-more-sense.md in the writing/substack/studio folder.`"

This is what I got back:

"A gritty tech-noir workbench featuring a modified retro-futuristic laptop running a holographic command-line interface. The hologram projects a glowing, complex directory tree structure of a digital studio, with amber text codscrolling. Background is a dim, cluttered workshop with blueprint schematics on the walls and scattered mechanical parts. Cinematiclighting, octane render, detailed engineering, used future aesthetic, 8k resolution." --aspect 21:9 --style gritty

Now I just paste that prompt into Nano Banana. Call that done. I am working on finishing the config for this. Once complete I will get set up to make it available for those interested. Getting back to Rails and AI being a match made in heaven.

Look React is awesome, but built for teams, not the solopreneur. Part of what makes React so amazing is the lack of opinion. There are so many ways to handle state, route, set up auth and nearly everything else. React was built that way. This means that it is up to you to make those decisions. If you don't have a technical history, this is certainly a challenge. There are platforms that have removed this complexity by either using NextJs or choosing for you. React is a frontend powerhouse, but are you building Figma or a Game?

React is Frontend magic. If you need backend magic, then Express is always a solid choice. AI tends to use NextJs for to address backend builds. If you need a database, Supabase is big in assisting AI with this these days. All of these technologies are great and very popular, but in one hand you get a bag of parts and note that says "good luck". In the other hand you are being forced to compromise so everything fits in your boxes.

Rails is solid and stable. It has been around since the 90's. The MVC architecture is tried and true, tested over time and corporations. Rails has better documentation than React. Rails is built on the most human readable software language. Built for minimum development to production time. The distinction between frontend and backend is clear. It is built to work with Postgres, out of the box. Run completely local on bare metal, on a cloud server, or in a docker container, out of the box. Stimulus and Hotwire give you SPA functionality built in. Rails will scale with the best of them. Rails 8 now includes a Turbo wrapper that effectively turns your PWA into a iPhone compatible app that can feature in the app store.

Remember when I asked if you were building a game? If you said no, then you have options. The real magic is in Rails and AI. Rails is magic on it's own, but LLM's write excellent Ruby because Ruby reads like English.

Like I said, it all seems a bit backwards and confusing these days. The times, they are changing. The AI industry feels like it is at "Sixes and Sevens" right now. I prefer a more blue collar approach. I am choosing order over this state of confusion and disarray.

6, 7 makes more sense