Agentic AI Foundations

Overview

How do agentic systems really work? Open the black box of an LLM, see how an agent is built on top of it, look at what agents can do today and learn where the risks are. In the hands-on part you build a small harness in Python and use a proxy to see every call between your harness and the model.

What you'll cover

• ▸Opening the black box: where LLMs sit within AI and machine learning, and what happens inside one, from tokens to the context window
• ▸From LLMs to agents: what a harness is, and how it runs the agent loop around the model
• ▸How a tool call works: how the model asks for one, and how the harness runs it and feeds the result back
• ▸How skills, MCP and memory give an agent extra instructions, tools and history, and where they end up in the context
• ▸What is currently possible: the main applications and the commercial landscape, from talk to your data to the SDLC and fully automated software teams
• ▸A word of caution: hallucinations, security and costs, when a fixed workflow is the better fit and what changes in production

What you'll be able to do

• ✓Explain how an LLM works and how an agent is built around it
• ✓Build a basic harness in Python that runs the agent loop with tool calling
• ✓Inspect the calls between a harness and a model through a proxy
• ✓Add a skill, an MCP server and simple memory to your harness
• ✓Map the main applications and the commercial AI landscape
• ✓Recognise the risks of agents and decide when a fixed workflow fits better than an autonomous agent