Appearance
Day 1 -- Foundations: How Models Read Prompts
Prompt engineering starts with understanding what happens between the moment you hit send and the moment a response comes back. Day 1 strips away the mystique: by the end of today you will know exactly how a model processes the input you give it, which parts of a prompt override others, and why the same technique that produces brilliant results in one situation produces garbage in another.
What you'll have by the end of today
- A working mental model of prompt engineering as a measurable, testable discipline -- not intuition, not "be more specific"
- The ability to dissect any prompt into its components and identify which parts are doing the work
- Hands-on experience with zero-shot, one-shot, and few-shot prompting, with code you can adapt to your own projects
- System prompt techniques that set a model's behavior, tone, and boundaries predictably
- A clear understanding of the instruction hierarchy -- which parts of a prompt win when they conflict
Lessons
- What Prompt Engineering Is -- distinguishes the discipline from folk wisdom, defines the engineering mindset (measurable, testable, optimizable), and surveys the techniques that make up the field
- Anatomy of a Prompt -- breaks down system prompts, user messages, and assistant prefills; covers instructions, context, input data, and output format, plus the primacy/recency effect in context windows
- Zero-Shot vs Few-Shot -- when to give examples and when they hurt; one-shot and few-shot mechanics, example selection strategies, and the bias problem nobody warns you about
- System Prompts & Role Assignment -- setting behavior, constraints, tone, and boundaries through system-level instructions; role-based prompting and the system/user separation
- Instruction Hierarchy -- which parts of a prompt override others when they conflict; why system beats user in most APIs, and how delimiter-based fencing and explicit priorities let you control precedence