Understand Claude
from scratch.

The simple definition

AI is software that learns patterns from data to make decisions or generate content. It doesn't think like a human — it recognizes statistical patterns in text, images, or numbers, and uses them to predict what comes next.

Types of AI you need to know

Claude = Anthropic's LLM

Claude is a large language model (LLM) built by Anthropic. Unlike other models, Claude was designed with an emphasis on being helpful, honest, and harmless — a philosophy called Constitutional AI. It has a massive context window (up to 200K tokens), which allows it to process entire books in a single conversation.

Claude model family (2025-2026)

What makes Claude different?

Tokens: the basic unit

An LLM doesn't read words — it reads tokens. A token is approximately 4 characters or ¾ of a word in English. When you send a message, it gets converted into a sequence of numbers (tokens). The model predicts the next most likely token, and that's how it generates text.

The inference cycle

Context window — working memory

The model only "remembers" what is inside its context window. Claude supports up to 200,000 tokens (~150,000 words). Everything outside that window simply doesn't exist for it.

The Claude Certified Architect – Foundations exam tests 5 knowledge areas. Here is each one explained in plain language, without exam jargon.

Agents — Claude that acts

An agent is Claude configured to make decisions in a loop: reason → act → observe the result → reason again. It can use tools, write files, call APIs, or delegate subtasks to other agents.

Tool Use — giving Claude hands

By default, Claude can only read and write text. With tool use, you give it access to real functions: search Google, query your database, execute code.

// Defines una herramienta
const tools = [{
  name: "buscar_cliente",
  description: "Busca un cliente por email en la DB",
  input_schema: {
    type: "object",
    properties: {
      email: { type: "string" }
    }
  }
}]

// Claude decide cuándo llamarla
// → "Voy a buscar al cliente con ese email..."
// → Tool call: buscar_cliente({ email: "juan@ejemplo.com" })
// → Recibe resultado → responde al usuario

MCP — Model Context Protocol

MCP is Anthropic's open standard for connecting Claude to any external system. Instead of defining tools by hand in every project, an MCP server exposes resources, tools, and prompts in a reusable way. It's like an API but designed specifically for LLMs.

Prompt Engineering

The way you talk to Claude determines the quality of its response. It's not magic — there are concrete techniques that significantly improve outputs.

// ❌ Prompt vago
"Analiza este código"

// ✅ Prompt estructurado
"Eres un senior engineer. Analiza este código Python
buscando: (1) bugs potenciales, (2) problemas de
rendimiento, (3) violaciones de PEP8.
Formato: lista numerada con severidad [HIGH/MED/LOW].
Código: [...]"

Multi-agent Systems

For complex tasks, instead of a single Claude doing everything, you can have multiple specialized agents working in parallel or in sequence. An orchestrator coordinates, while workers execute specific tasks.

Orchestrator (Claude Opus):
  → Recibe: "Analiza el mercado de EVs en España"
  → Descompone en 3 subtareas paralelas:

  Worker A: Scraping noticias últimos 30 días
  Worker B: Análisis de datos de ventas
  Worker C: Investigación de competidores

  ← Recibe resultados de A, B, C
  ← Sintetiza informe final

Starting from zero knowledge about Claude. Spending ~1 hour per day. The exam is realistic with 4-6 weeks of consistent practice.