Every agent is built around a large language model (LLM), so it helps to know what's happening inside that box. At its core, an LLM predicts the next piece of text, one token at a time, using patterns learned from huge amounts of data. This module covers how those models are trained, what they're good and bad at, and why the context window matters so much for agents.
The first two lessons build that mental model — a short animated explainer, then a longer talk on how LLMs are trained, used, and attacked. The last lesson turns to prompting: writing clear instructions, giving examples, and iterating. These skills carry straight into agent work, where the system prompt is how you tell an agent who it is, what it can do, and how it should behave.
Watch the lessons in order. The last two are long, so treat them as references you can come back to as you build agents in later modules.
3Blue1Brown's short animated introduction to what an LLM is: a function that predicts the next word, how pre-training on huge text collections tunes billions of parameters, and where transformers fit in.
Video 8minAndrej Karpathy's jargon-light talk on how LLMs are trained and fine-tuned into assistants, how they use tools, and security issues like jailbreaks and prompt injection that matter once agents can act.
Video 1hA roundtable of Anthropic's prompt engineering experts on what makes a good prompt: clear communication, iterating and testing, reasoning techniques, and how prompting practices are changing.
Video 1h 16min