The AI in AI automation is usually a large language model placed as one step inside a workflow. It reads some text, such as an email, a form answer, or a support ticket, and produces some text back. To use it well, you need a basic picture of what it is doing, because a model that predicts the next word behaves very differently from a traditional piece of software.
Most of your results come from the prompt. A clear instruction, a defined role, and a few examples of the output you want will do more than any setting. A workflow also needs answers it can use, not paragraphs of prose, so you will learn to ask for structured output, usually JSON that matches a fixed shape, so the next step can read each field reliably.
Finally, function calling (also called tool calling) lets a model choose from a list of actions you describe and return the arguments needed to run one, which is the foundation for the agents you will build later in the course. Watch the lessons in order, then continue on to Data & Integrations.
3Blue1Brown's short visual introduction: an LLM as a function that predicts the next word, how it is trained, and why transformers made large models practical.
freeCodeCamp's walkthrough of prompt engineering, from how language models work to the prompt engineering mindset, zero-shot and few-shot prompts, and best practices.
A conference talk on getting reliable structured data out of LLMs: defining the output as a schema, validating it, and letting the model retry when the result is wrong. It uses Python, so treat it as a look at the idea behind structured output.
A hands-on function calling example in which an OpenAI model decides, from the conversation, when to call a function you have defined, in this case ending a call.