{"id":35699,"date":"2026-04-29T13:50:58","date_gmt":"2026-04-29T11:50:58","guid":{"rendered":"https:\/\/www.codemotion.com\/magazine\/?p=35699"},"modified":"2026-04-29T13:51:01","modified_gmt":"2026-04-29T11:51:01","slug":"ai-stack-from-machine-learning-to-agentic-ai","status":"publish","type":"post","link":"https:\/\/www.codemotion.com\/magazine\/ai-ml\/ai-stack-from-machine-learning-to-agentic-ai\/","title":{"rendered":"AI Stack: From Machine Learning to Agentic AI"},"content":{"rendered":"\n
\n

A practical guide for developers who want to understand the AI ecosystem \u2014 not just use it.<\/p>\n<\/blockquote>\n\n\n\n

\n\n\n\n

The word “artificial intelligence” gets applied to everything: chatbots, Netflix recommendations, self-driving cars, image generators. But behind all of it sits a well-defined technology stack \u2014 layers built on top of each other, each one with a reason to exist. Understanding this structure is what separates developers who use AI tools from developers who design them.<\/p>\n\n\n\n

Let’s start from the bottom.<\/p>\n\n\n\n

\n\n\n\n

1. Artificial Intelligence (AI) \u2014 The Umbrella<\/h2>\n\n\n\n

AI is the field that encompasses everything: any system designed to simulate human cognitive capabilities like reasoning, learning, perception, or decision-making.<\/p>\n\n\n\n

The earliest AI systems were rule-based<\/strong>: explicit logic written by hand.<\/p>\n\n\n

def approve_loan(income, credit_score):\n    if<\/span> income > 50000<\/span> and credit_score > 700<\/span>:\n        return<\/span> \"Approved\"<\/span>\n    return<\/span> \"Denied\"<\/span>\n<\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
Simple, transparent, brittle. It doesn’t adapt. It doesn’t learn. Every new edge case requires a human to update the rules manually. That limitation is exactly what motivated the next layer.<\/p>\n\n\n\n
\n\n\n\n
2. Machine Learning (ML) \u2014 Learning from Data<\/h2>\n\n\n\n
ML flips the paradigm: instead of writing rules, you provide data<\/strong> and let the model learn the patterns on its own.<\/p>\n\n\n
from<\/span> sklearn.datasets import<\/span> load_iris\nfrom<\/span> sklearn.model_selection import<\/span> train_test_split\nfrom<\/span> sklearn.ensemble import<\/span> RandomForestClassifier\n\nX, y = load_iris(return_X_y=True)\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2<\/span>)\n\nmodel = RandomForestClassifier()\nmodel.fit(X_train, y_train)\n\npredictions = model.predict(X_test)\nprint(predictions[:5<\/span>])\n<\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
Where it shines:<\/strong> fraud detection, demand forecasting, recommendation systems, churn prediction.<\/p>\n\n\n\n
Where it shows its limits:<\/strong> high-dimensional data, images, text, audio. For those cases you need more powerful architectures.<\/p>\n\n\n\n
\n\n\n\n
3. Neural Networks (NN) \u2014 The Artificial Brain<\/h2>\n\n\n\n
Neural networks take inspiration from the structure of the brain: interconnected nodes that process information in parallel. Each neuron computes:<\/p>\n\n\n
output = activation(weight \u00b7 input + bias)\n<\/code><\/span><\/pre>\n\n\n
Training happens through backpropagation<\/strong>: compute the error on the output, propagate it backward, and update the weights to minimize it.<\/p>\n\n\n
import torch\nimport torch.nn as nn\n\nclass NeuralNetwork(nn.Module):\n    def __init__(self):\n        super().__init__()\n        self.layer1 = nn.Linear(4, 8)\n        self.layer2 = nn.Linear(8, 3)\n\n    def forward(self, x):\n        x = torch.relu(self.layer1(x))\n        return self.layer2(x)\n\nmodel = NeuralNetwork()\noutput = model(torch.rand(1, 4))\nprint(output)\n<\/code><\/span><\/pre>\n\n\n
Neural networks work well on structured problems. But for genuinely complex tasks \u2014 speech recognition, language understanding \u2014 you need networks that are much, much deeper.<\/p>\n\n\n\n
\n\n\n\n
4. Deep Learning (DL) \u2014 More Layers, More Power<\/h2>\n\n\n\n
Deep Learning is ML with many-layered neural networks. More layers means the ability to extract increasingly abstract features from raw data.<\/p>\n\n\n\n
This is what made computer vision, NLP, speech synthesis, and autonomous driving possible.<\/p>\n\n\n\n
The main architectures you’ll encounter in practice:<\/p>\n\n\n\n
Architecture<\/th>Typical use case<\/th><\/tr><\/thead>
CNN<\/td>Computer vision, image classification<\/td><\/tr>
RNN \/ LSTM<\/td>Time series, sequential text<\/td><\/tr>
Transformer<\/td>NLP, LLMs, generation<\/td><\/tr>
Autoencoder<\/td>Compression, anomaly detection<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n
from<\/span> tensorflow import<\/span> keras\nfrom<\/span> tensorflow.keras import<\/span> layers\n\nmodel = keras.Sequential([\n    layers.Dense(128<\/span>, activation=\"relu\"<\/span>),\n    layers.Dense(64<\/span>, activation=\"relu\"<\/span>),\n    layers.Dense(10<\/span>, activation=\"softmax\"<\/span>)\n])\n\nmodel.compile(\n    optimizer=\"adam\"<\/span>,\n    loss=\"sparse_categorical_crossentropy\"<\/span>,\n    metrics=[\"accuracy\"<\/span>]\n)\n\nmodel.summary()\n<\/code><\/span>Code language:<\/span> JavaScript<\/span> (<\/span>javascript<\/span>)<\/span><\/small><\/pre>\n\n\n
Deep Learning opened the door to generating new content \u2014 and that’s where things get really interesting.<\/p>\n\n\n\n
\n\n\n\n
5. Generative AI \u2014 Create, Don’t Just Classify<\/h2>\n\n\n\n
A classifier says “this image is a cat.” A generative model creates a new image of a cat<\/strong>. That’s a fundamental paradigm shift.<\/p>\n\n\n\n
The main families of generative models:<\/p>\n\n\n\n