AI for Noobs

Published: 2026-02-03

Intro

It's February in the year of our lord 2026. 3 years after the release of OpenAI's Chat GPT2.0 LLM (and trillions of dollars), AI hype is at fever pitch. We've got LLM's, RAG's and MCP's... OH MY! It turns out, AI is better when you don't have skills issues.

If you are confused, welcome to the club. This post is my attempt to understand and keep track of what is happening with the tooling in this space.

LLM

Large Language Models (LLM) are the foundational technology behind the AI craze. LLMs are trained on very `large` datasets (often in the thousands to millions of Gigabytes. LLMs are then fine tuned to towards specific tasks such as language translation, question answering and code generation.

Machine Learning

Transformers

Pre-Training

Labelling

Weights

MCP

MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems.

MCP defines a well known standard to allow bi-directional communication channels between AI applications such as Claude Code or Copilot-CLI and external systems such as data sources like Netbox or SaaS tools such as Slack.

Examples

Tools

Tools are functions that are registered with an LLM via a schema that can be used to retrieve or process data

Examples

Agents

Agents are autonomous sotware processes that can plan, make descisions and complete tasks independently or with minimal human intervention.

Planning

Decision Making

Reasoning

Goal Orientated

Learning

Tool Usage

Examples

Skills

Skills are packages of domain specific knowledge that influence how agents reason about and approach solving tasks. Skills do not execute any code directly, they inform the agent decision making about a specific domain.

Because skills have domain specific knowledge, they can be more efficient with token usage.

Skills skills.sh

Examples

Tokens

Tokens

Context

Context allows you to give the LLM additional information to help it achieve your goal. For example, context can be source code, an error message, an image or a PDF, even a link to a webpage.

LLM's will hold the context you provide (as well as the full conversation which gets used as context) in it's Context Window. The context window is not infintely large. Therefore, the important thing about context, is to provide relevant context related to completing the task, without overloading the context window.

When the context windows gets full, the LLM will summarise the current window and compact it so that it can contiue working. In this process you lose alot of the existing context.

ReAct

Agentic RAG

Agent-to-Agent

Agent-to-Agent

RAG

Retrieval-Augmented Generation

LangChain

LangChain

LangGraph

LangGraph

People

  • Boris Cherny - Anthropic/ClaudeCode

Outro

Tags

#ai