Introduction to Vibe Coding with Agentic IDEs

This section is considered the core of the course, as we will thoroughly explore the software's essential functionalities that enable us to program with AI assistance.

By the end of this section, students will have acquired the practical knowledge necessary to start "Vibe Coding" effectively.

What you will learn in this section:

• Basic Prompt Engineering: How to structure and write an initial prompt correctly.

• Iteration and Refinement: How to request specific changes to the generated code to adjust it to your needs ("Making changes to the game").

• Debugging with AI: Techniques to identify and fix errors (Fixing bugs) using the agent itself.

• Multimodal Prompts: How to enhance your instructions by combining text and image to provide visual context to the AI.

• Voice Interaction: How to streamline the workflow by entering prompts (both text and image) using voice commands.

In this practical lesson, we will perform our first "Vibe Coding" exercise. You will learn how to navigate the Windsurf interface, select the appropriate AI model (such as Claude Opus or Sonnet), and use the Cascade agent to generate a complete web application.

We will see step by step how to:

• Initialize an empty project.

• Ask the AI to create the classic "Snake" game.

• Automatically generate and accept HTML, CSS, and JavaScript files.

• Run and test the video game in the browser instantly.

  
  

Once we have the functional base, it is time to bring it to life. In this lesson, we will learn how to iterate on an existing project using natural language instructions.

We will see how to ask Windsurf's Cascade agent to make complex changes without manually touching a single line of code:

• User Experience: We will add sound effects and a particle animation (explosion) every time the snake eats.

• Code Review: You will learn to interpret, accept, or reject the code blocks (diffs) that the AI proposes in specific files like game.js.

• Design Adjustments: We will adjust the game area and snake sizes to improve gameplay, demonstrating the flexibility of AI-assisted development.

In this lesson, you will learn how to manage the inevitable setbacks in software development and how to use Artificial Intelligence to recover quickly. Code doesn't always work on the first try, and sometimes the ideas we test don't turn out as we expected.

In this lesson, you will learn a fundamental Vibe Coding technique to prevent Artificial Intelligence from generating generic or incomplete code: technical specificity through layered prompts.

Instead of directly asking the editor to "make a game", we will see a two-step workflow to replicate a classic video game (Snake) with high fidelity:

• Generation of the specification: We will use ChatGPT to analyze a visual reference and convert it into an extremely detailed technical prompt (describing mechanics, logic, visual style, and specific technologies like HTML5 Canvas and JavaScript).

• Execution with Agents: We will feed that detailed description to Windsurf (Cascade) to generate the complete project in one go.

You will discover why "the more details you give the prompt, the better the model works", ensuring the agent understands exactly the behavior, colors, and game rules without the need for tedious iterations.

What you will see in this video:

• How to use an LLM to write prompts for another LLM.

• The importance of defining the technology stack (JS + Canvas) from the start.

• Real-time code generation with Windsurf.

• Final test of the functional game with score, sound, and restart.

In this lesson, you will learn how to use the multimodality of the most advanced AI models. We will see how a simple screenshot can replace a long text description when programming.

What you will learn:

• How to pass an image as context (prompt) to your AI agent (Windsurf/Claude).

• The AI's ability to recognize visual styles, colors, and mechanics from a photo.

• Quick comparison: Which models support vision (Claude, GPT-4, Gemini) and which do not (DeepSeek, Grok).

• Practical case: Functional and visual recreation of the "Snake" game based solely on a reference image.

In this class, we will explore the two most powerful alternatives to Windsurf for performing Vibe Coding today.

We will analyze Cursor, consolidated as the industry standard thanks to its high levels of privacy, and I will introduce you to Antigravity, Google's new agentic editor that is quickly gaining popularity due to its integration into the company's ecosystem.

Discover which of these tools best fits your AI-assisted development workflow.

In this lesson, we explore Cursor, one of the most powerful and popular tools in the "Vibe Coding" ecosystem and Windsurf's main competitor. Through a practical and fun exercise, we will see how this editor (a VS Code fork) can drastically accelerate your workflow.

What you will learn in this class:

• Initial setup: Download, installation, and project creation in Cursor.

• Interaction modes: Differences between Chat, Plan, and Agent modes, and how to choose the right model (in this case, using the free version of Grok).

• Voice programming: How to use audio commands to dictate complex instructions and generate the complete structure of a project.

• Creation of a real project: We will develop a clone of the classic Breakout game (Atari) from scratch using web technologies (HTML, CSS, and JavaScript).

By the end, you will have a functional video game with sound effects and scoring, created entirely through prompts and corrections in natural language.

Is it worth switching to Google's new IDE?

We analyze Antigravity, a tool that shares DNA with Windsurf but takes automation a step further through deep integration with Google services.

In this video, we will see:

• The similarities and improvements compared to other AI editors.

• The "Vibe Coding" workflow is applied to a classic video game project.

• How the agent manages errors and performs automatic corrections.

  
  

In this video, I present the three practical challenges we will address in this section to master Vibe Coding in Python. We will move from theory to practice by building:

• Automated Data Analyst: A system that ingests Excel files, finds patterns, and generates charts automatically.

• Computer Vision (Classification): An AI capable of analyzing MRI scans to determine the presence of brain tumors.

• Advanced Detection (Localization): A system inspired by YOLO that not only detects but also locates and frames tumors in CT scans.

Nothing kills a "vibe" faster than a project that suddenly stops working because of a software update. In this lesson, we establish the most critical foundation for Python development: Isolation. Before we start generating code with AI, we need to ensure our workspace is bulletproof.

What we will cover:

• The "System Python" Danger: Why you should never touch the Python version pre-installed on your operating system (macOS/Linux) and the risks of "Dependency Hell."

• The Container Concept: Understanding how Virtual Environments act as self-contained "universes" for each of your projects.

• Why Anaconda?: We will introduce Anaconda as our tool of choice for managing these environments, ensuring that the libraries we use for Data Science or AI today won't break our projects tomorrow.

Why this matters for Vibe Coding:

When you ask an AI Agent (like Windsurf or Cursor) to write code, it uses specific library versions. By using virtual environments, you ensure that the code the AI writes for you remains stable, reproducible, and bug-free, regardless of future updates to the Python language.

In this video, you will learn how to install Anaconda in Windows and how to create a new virtual environment for our Python projects.

Before we let our AI agents loose to write code, we need to build them a home. In the world of Vibe Coding, we prioritize speed and experimentation, but that requires a clean, organized workspace to prevent chaos.

In this lesson, we will install the "brain" behind our operations: Anaconda (Miniconda). You will learn how to create isolated Virtual Environments—essentially "sandboxes" where we can install libraries and test code safely. This is a critical skill; it ensures that when we ask the AI to build a tumor detector or a data analyzer, it doesn't conflict with other projects on your machine.

What we will cover:

• Miniconda Installation: How to set up the lightweight version of Anaconda on macOS (Apple Silicon).

• Terminal Basics: Verifying your installation using the command line.

• The "Sandbox" Strategy: Using conda create to build a dedicated environment named vibe_coding.

• Activation: How to switch from your computer's "base" environment to your new project space using conda activate.

By the end of this video, your Mac will be ready to speak Python 3.11, and you'll have the infrastructure ready for the advanced applications we are about to build.

How do experts transform data into decisions?

In this section, we apply everything we have learned about Python to build a real analysis pipeline. Through Vibe Coding, you will see that you don't need to be a mathematician to understand your data; you just need to know how to ask the code the right questions.

You will learn to:

• Load CSV/Excel files and prepare them for analysis.

• Execute quick queries to filter relevant information.

• Interpret the results to make data-driven decisions.

In this lesson, we will carry out an "End-to-End" project focused on medical computer vision. The workflow will be as follows:

• Data Acquisition: We will download a real MRI dataset directly from Kaggle.

• Training: We will create and train a binary classification model capable of distinguishing between healthy brains and those with tumors.

• Deployment (Web UI): We won't stop at code alone; we will build an interactive web interface. You will be able to upload a test image and see the prediction and your model's confidence level in real time.

Sometimes, a model with 99% accuracy is too good to be true. In this critical lesson, we move beyond simple training and dive into Error Analysis, the art of understanding why a model fails.

We will encounter a common but dangerous pitfall in Machine Learning: Data Leakage. You will see how our model was "cheating" by memorizing patient anatomy rather than detecting tumors. Instead of spending hours manually auditing thousands of files, we will use Vibe Coding to have our AI agent investigate the dataset structure, discover the leakage pattern, and rewrite our data-splitting logic from scratch.

In this video, you will learn:

• Visual Error Analysis: How to generate a visual dashboard to identify "False Negatives" and "False Positives" that don't make sense.

• Detecting Data Leakage: Understanding the difference between Image-level splitting vs. Patient-level splitting and why getting it wrong ruins your model.

• AI Forensics: How to prompt the AI to analyze file naming conventions (e.g., Te-no, Tr-no) to confirm if training data is bleeding into the test set.

• Patient-Aware Splitting: Using the agent to write a robust Python script that ensures a clean separation of data, guaranteeing your model's accuracy is real.

• Retraining & Verification: running the fixed pipeline to achieve authentic, high-confidence results (99%+) on completely unseen data.

Your model has 99% accuracy, but… do you trust it? In the world of Deep Learning, we often work with 'Black Boxes': we know they work, but we don't know how.

In this lesson, we will take Vibe Coding to the next level. We won't settle for cold metrics; we'll visualize the AI's thinking. Using Explainable AI (XAI) techniques and the Grad-CAM library, we will ask our code assistant to develop a complete web interface.

You will see, step by step, how to generate heatmaps that reveal exactly which pixels your neural network is looking at to detect tumors (or any other object). Together, we will discover if your AI really learned to identify the pathology or if it is just looking at the image background.

Get ready to audit your models like a pro!

In the previous lesson, we achieved incredible accuracy in image classification, but the real world demands more. It is not enough to know if there is a tumor; we need to know exactly where it is.

In this video, we will test the limits of Vibe Coding. We will leave simple classification behind to delve into Object Detection using one of the fastest and most modern architectures in the world: YOLOv10.

You will see in real time how I go from a complex idea to a functional implementation without manually writing training code. We will use Windsurf and its Cascade agent to orchestrate the entire process.

What you will discover in this session:

• From "What" to "Where": You will understand the critical difference between image classification and detection with bounding boxes.

• Real Data Management: How to search, download, and clean complex Kaggle datasets and prepare them for the demanding YOLO format.

• The Vibe Coding Flow: You will see me make mistakes, face agent blocks, and solve virtual environment and library issues live. Vibe Coding is not magic; it is knowing how to guide the AI when it stumbles.

• YOLOv10 Training: We will configure and train a state-of-the-art neural network on a Mac (CPU), demonstrating that you don't always need a GPU farm to start prototyping.

This lesson is the ultimate proof that with the right tools and the right "vibe", what used to take weeks of development is now done in the time it takes to drink a coffee.

Your model reports 99% accuracy, but when you run the app... it fails completely. Welcome to the reality of AI development.

In this crucial session, we face a "silent failure" scenario where our Tumor Detection System looks perfect on paper but fails in production. Instead of guessing blindly, you will learn how to use Vibe Coding to orchestrate a forensic investigation of your model.

What you will experience in this lesson:

• The "Sanity Check" Workflow: How to instruct your AI agent to create custom visualization scripts that compare Ground Truth vs. Predictions side-by-side, revealing the truth behind the numbers.

• Hardware vs. Software: Watch as the AI agent digs deep to discover a rare, obscure bug involving Apple Silicon (MPS) tensor corruption—a complex issue that would take a human developer hours to isolate, found and fixed by the agent in minutes.

• Iterative Repair: We won't just fix the code; we'll fix the system. You'll see how to guide the agent to patch the inference logic in the web app without breaking the rest of the stack.

• The Final Polish: We conclude by deploying the fixed YOLOv11m model into our Gradio interface, achieving a fully functional, professional-grade medical diagnostic tool.

This lesson proves that Vibe Coding isn't just about writing code faster; it's about having a senior engineer pair-programmer who can help you solve the bugs that usually kill projects.

In this introduction, we will explore the technology stack we will use to create native applications. Discover how we will combine Android Studio for project management, ChatGPT for prompt design, and Windsurf (or Cursor) for code generation, achieving an accelerated and efficient development workflow.

Do you think you need to be a Kotlin expert or know complex programming languages to create your first App? Spoiler: No. In the world of Vibe Coding, technology works for us.

In this fundamental lesson, we will get everything ready for the magic to happen. Not only will we install the necessary software, but I will also teach you how to configure your most important tool: your own smartphone.

What we will do step by step:

• Download and Installation of Android Studio: We will see how to obtain the official development environment (IDE) and perform a clean installation using the recommended settings, without complications.

• Goodbye to traditional programming: You will understand why in this course we will not focus on code syntax, but on creation.

• The "7-Tap Trick": I will teach you how to unlock the hidden Developer Options on your Android phone.

• USB Debugging: We will activate the bridge connection that will allow the apps you create on your computer to travel "over the air" (or by cable) directly to your cell phone to test them in real time.

By the end of this video, you will have your "ship" ready to take off and start creating native applications without writing a single line of complex code. Let's get to work!

In this lesson, you will take Vibe Coding to the Android world by creating a Pac-Man-style game in a practical and fun way. You will learn how to transform an idea into a functional app by connecting game logic, state management, animations, and touch controls, all with a creative, fluid mindset.

It is not just about writing code, but about feeling the flow of development, experimenting quickly, and seeing immediate results on your own Android device. By the end, you will have a solid foundation for creating games or interactive apps and will understand how to apply Vibe Coding to real projects in Android Studio.

Ideal for developers who want to learn by doing, enjoy the process, and build something visual and playable from the very first moment.

Congratulations! You’ve reached the end of the course, but this is only the starting point of your journey as a Vibe Coder.

In this final message, we reflect on the rapid evolution of Agentic Development Environments like Windsurf, Cursor, and Antigravity. As we've seen, these tools are constantly being updated with new features that redefine how we write software.

What we cover in this closing session:

• The "Iceberg" of AI Tools: Understanding that we've covered the essentials, but each of these IDEs has a depth of features that could fill entire courses on their own.

• Embracing Change: How to stay adaptable in an ecosystem where the software updates faster than the documentation.

• The Ultimate Takeaway: Realizing that your technical limitations are gone. Whether it's a complex app or a simple script, you are now just one well-crafted prompt away from your goal.

Thank you for taking this journey with me. Now, go let your imagination run wild and build something amazing!