Vibe Coding Prototypes Agentic Development With Ai In Edunit

The Two Paths of AI Building: Vibe Coding vs. Agentic Coding A clear breakdown of different approaches to building with AI assistance In this article I'd like to explore two concepts: Vibe coding and agentic coding. Placing tools like Lovable, Bolt and tempo on the vibe coding side and Cursor, Claude Code and Windsurf on the Agentic Coding end. Since these are relatively new concepts, there's a lot of misunderstanding and false promises about what each can achieve.

I'll try to make it digestible for anyone with absolutely no prior tech experience. The goal is understanding the most effective path for using AI to build your idea and ideally monetize it. I recognize that vibe coding uses an agent to complete tasks, yet throughout this article you'll understand why it feels appropriate to make clear distinctions from a utility standpoint as well as regarding the target audience for each. AI for Building Before anything else, let's understand what's behind this hype around AI and development.

How did we suddenly have absolute beginners building entire websites? To keep it simple: LLMs (Large Language Models) They are language models in the sense of predicting which characters (tokens actually) comes next. We've trained these models with lots and lots of code as well.

So whenever you ask AI (even inside ChatGPT) to "create a login page for me," it will find the most generic implementation of that and send you the code: It created the code using a very popular stack among developers, yet people have no idea how to run this code. The workaround most have to this is going back to the basics: HTML.

Some people already understand some core concepts and know an HTML page is as simple as saving a .html file and opening it, so the prompt becomes "create a login page for me in pure html." The result: Beautiful! But we need a register page, a landing page, the authentication setup, a database, we also need to deploy the app and much more. Most of all, we need iterations when building something.

And doing it all with ChatGPT isn't very efficient, nor is placing the entire code in a single HTML file. Why? Well, the iterations would basically happen like this: Eventually the HTML code would be so huge we'd run into either of these problems: Too expensive to continue iterating. No more context window (AI memory) A solution is working with components. Instead of having the AI fetch a single code file, we'd only provide it the specific part of the code we need built.

So now we can use the context window to build specific parts of the app. In the example above, it would use the register page example to be able to build the login page. But if we eventually want to build the landing page, we can just open a new chat (context window) for that, since the AI doesn't have to know the business logic for authentication while building the landing page. Yet... if you don't have a tech background, you're probably already getting confused. But we've made progress.!

You've seen code up to this point, but you already understand you don't have to code a single line to have something built, correct? What's missing? Structure and an agentic capability. Calm down! I'll make it easy to understand. Structure has to do with how our project organizes code. Developers deal with this daily to find code in a huge codebase.

Example: Pages > Auth > - Login - Register If this is my code structure, I know that to access my register page code I'll go inside the "pages" folder, then the "auth" folder and find the "register" code in there. This is where it gets interesting with AI: Agents. An agent is just an AI with tools and a loop for a given task.

The logic is more or less like this: User inputs: "Edit the register button, make it blue." AI thinks: "I need to make the register button blue. But where the f* am I?" AI executes tool: tool reads directories (AI sees the folder names and some file names) AI thinks: "Oh, I seem to be inside of a web app project in a NextJS framework.

From my training I know that pages are usually placed inside the 'pages' folder, so let's check if there is a 'register' page in there." AI executes tool: tool goes inside the "pages" folder AI thinks: "Hmm, there's an auth folder in here, let's see what's in that." AI executes tool: tool goes inside the "auth" folder AI thinks: "Aha, here's the register page, let me find the button." AI executes tool: tool reads maximum of 100 first lines of the code AI thinks: "Didn't find the button, let me keep searching." AI executes tool: tool reads next 100 lines of code AI thinks: "Found it!

Let's make it blue." AI executes tool: tool changes code and adds a blue background to the button AI thinks: "Task done!" End. So now we don't have to worry about finding the right context, we let AI do that for us. Yet still, to get to this point you'd already be running the project. What if you needed to start fresh? That's where we get into the "vibe-coding" turning point.

Vibe Coding The term vibe-coding started with this X post: He clearly specifies "forget that the code even exists." I mean... in my opinion he didn't mean that in such a brain-dead way, but a lot of companies saw an opportunity with this. A non-technical person doesn't know how to start building a website. Most people don't know the difference between backend and frontend. They don't know how a database works, let alone how to configure authentication or payments. So the simple idea was having all this pre-configured...

all the code strucure in the AI's context so it knows exactly how to work with it... and finally have everything already deployed. Let's compare this with a Subway sandwich. The bun is the foundation, the setup. The "structure" we talked about previously, all pre-configured along with the deployment. The fillings are whatever the user wants to build. They just ask and it's there, all integrated.

Now we have the perfect "easy" and "fast" approach to building something people have known for nearly a decade makes quite a lot of money: web apps. Here's where it gets tricky: Most developers were bothered by this. I mean, imagine you've been building apps for years and suddenly there are claims of people with no prior coding experience making hundreds of thousands of dollars with their AI-built app. Having worked as a developer and learning how other developers see projects, I made up my own theory about this...

Engineers like solving problems. Often we are attracted by the technical difficulty behind a project, which is absolute dopamine for us. It's like solving a Rubik's cube or a puzzle... such that when we solve the technical part which was basically 80% of the app and took us a week, the other 20% might take months and will likely be abandoned. So the correlation between a successful web-app and a genius tech implementation is made automatically. Therefore making developers skeptical about people actually making money with vibe coding.

In reality, niche expertise is everything. If you know your niche well enough, you know exactly what bothers you and how that could be fixed and even how you'd distribute the idea—that's absolute gold. So the technical implementation most of the time is just the absolute basic front-end input and output, but it solves a problem, it brings value and therefore makes money. I just hate the preaching about these vibe-coded apps being scalable and robust. They aren't. Most don't make money, look ugly, are unusable and have zero value.

Most, not all of them. I just guess that since the marketing attracts the "make money fast and easy" audience, they're just too lazy to even try to understand the limitations of AI web app builders. And even so, we have a bunch of high-tech websites that look amazing, are efficiently usable, receive hundreds of thousands in funding, with an entire dev team and still provide zero value. So yeah, vibe-coding basically is using AI to democratize micro-SaaS builds. They validate your idea. This alone is MASSIVE.

But there's an alternative that I'd say is the real gold in using AI to build web apps: Agentic Coding. You've probably heard about "Cursor" or "Claude Code" or "Windsurf." Well, I wouldn't place these tools inside the "vibe-coding" category for a couple of reasons. Agentic Coding Different from vibe coding, "Agentic Coding" has no specific person, paper, or moment that it started to be used. I discovered this term watching one of my favorite content creators: web dev cody.

Since I didn't see myself in the "vibe-coding" circle, but understood perfectly how much AI helps us build things, this term seemed perfect. My personal experience might matter a bit while explaining my enthusiasm with Agentic Coding, so here it is: I'm a full-stack developer, and built websites in PHP nearly 10 years ago... my first successful app was exactly what I explained above: zero tech skills, just delivering what people needed (it was vanilla JavaScript with pure CSS and a lot of Stack Overflow copy/paste code).

But with an engineer brain I had to learn more and create solid apps, so I dove into the modern full-stack ecosystem of building apps the fastest way possible: Next.js, Tailwind, TypeScript, shadcn, Prisma, EasyPanel and much more. At this point I'd understood the exact lines of code I needed to type to make a Next.js page render statically, or to ensure my API routes were protected. I'd build the same structure hundreds of times. But as AI models evolved I noticed something interesting: I wasn't coding anymore.

I didn't have to look up how to perform incremental static regeneration in Next.js > write the code > build the app and see if it was rendered statically. Now I was just asking the AI "make the page X perform incremental static regeneration, revalidate every 1 hour." It would code everything, build the app, check if it was successful and... done. This made me extremely curious to see if someone with absolutely no coding experience, but the right concepts in mind would be able to create an entire app.

So I tested it with some members of my Skool community as well as with some friends that were interested in becoming AI automators. The first thing I had in mind was: the AI has to have FULL context of the entire app, so all the tools need to be integrated in the codebase. I'd also explain to them that Next.js handles front end and back end, I'd explain why to use shadcn and how Prisma would allow the AI model to apply changes to our database independently. And...

they did it. Jorge, with absolutely zero coding experience was able to build a Chrome extension that synchronizes a sign language interpreter video with Netflix content in real-time. He's already contacting advertisers to sponsor his project and has an audience just waiting for the final launch. Agentic coding allowed him to explore more than web apps. No, I still wouldn't market this as "build a scalable web app in days with AI." Without the coding expertise and enough context, I don't believe current LLMs can build that.

But I've noticed much more flexibility for these MVPs than those built with vibe coding apps. It's like if in vibe-coding you always have to go with the white bread they bake for you while also risking asking for an additional option from the menu that you thought would be free. While in agentic coding you're at your own house with the bread of your choice that can either be a web app, a Chrome extension, a Python script to track prices, a bot for your RPG game, really anything.

But you'll have to bake it. Vibe coding apps get you better build results initially, but these build quality will improve along with your ability to provide better context and your general concept knowledge. While with agentic coding you'll likely take a bit longer to improve, which is basically the baking phase of setting up the project and deploying it. Once you nail this, the build quality should improve exponentially.

Both options are viable, but one is a further step in the direction of building a reliable cost-efficient app: To sum everything up: Use vibe-coding to understand how AI helps us build things. Learn basic concepts and tools like: backend, front end, full-stack, Next.js, Postgres, UI, shadcn, Prisma. Move to agentic coding and use Claude Code / Cursor to build your app. Don't get me wrong. Vibe coding tools are improving like crazy. The speed, the polish, the features they're adding... it's impressive stuff. The problem is everyone's using them.

You've got thousands of people launching similar-looking apps every week. If you want to actually differentiate yourself from this massive wave of vibe-coded apps, you're honestly just a handful of tools, some setup knowledge, and basic concepts away from stepping into a completely different league. We're talking weeks of learning, not years. And while I can't promise you'll build the next unicorn SaaS, you'll absolutely be able to create efficient single-feature micro-SaaS products without touching code.

That's exactly what I'm testing with the 10-day course I just launched in The AI Forge. I want to see if this theory holds up in practice. Can people make this transition as smoothly as I think they can? I'll be sharing the results and member builds right here in future newsletters. Honestly can't wait to see what happens.

People Also Asked

• Vibe Coding vs. Agentic Coding: 2025 Beginner's Guide to AI‑Driven ...
• The Two Paths of AI Building: Vibe Coding vs. Agentic Coding
• Vibe Coding vs. Agentic Coding: Fundamentals and Practical Implications ...
• The Future of AI-Driven Development: Vibe Coding vs. Agentic Coding
• From vibe coding to agentic AI: A roadmap for technical leaders
• Agentic Engineering: The Complete Guide to AI-First Software ...
• Agentic Engineering: The Post-Vibe-Coding Paradigm - Morph