How LLMs & Agents Work

Days 1–2 · no prerequisites

Before using these tools well, it helps to have an accurate mental model of what they are — and what they are not. These first days build that foundation.

Large language models

• What a language model is, and what "predicting the next token" does and doesn't imply.
• Context windows: what the model can see, and why what you put in front of it matters so much.
• Strengths and failure modes — including why a confident answer can still be wrong, and what "AI slop" looks like.

From models to agents

• What makes a system agentic: tools, actions, observations, and a loop.
• The difference between a single prompt, a chat assistant, and an agent that takes actions in the world.
• Why agency introduces new risks alongside new capabilities.

Quality vs. slop

A recurring theme of the whole course starts here: telling the difference between genuine quality and plausible-looking output, and learning to engineer for quality rather than hoping for it.

Outcomes

After these two days you should have a mental model accurate enough to predict, roughly, when these tools will help and when they will mislead — the foundation for everything that follows.