Learning path · Foundations · 04
Prompting
Structuring instructions, context, and examples so the model produces useful, controllable outputs for a specific task.
Why it matters
- Prompt quality often moves outcomes more than swapping model families.
- Clear roles, constraints, and formats reduce ambiguity and rework.
- Prompting is the first control surface before orchestration gets complex.
Key ideas
- Task specification
- Constraints
- Format control
- Iteration
Top resources
- 01DocsOpenAI
Prompt engineering
Why this resource. Practical control surface: roles, constraints, and iteration before you reach for agents.
Covers in this concept
- task specification
- constraints
- format control
- 02DocsAnthropic
Prompt engineering overview
Why this resource. Complementary prompting patterns with a stronger emphasis on XML structure and evaluation.
Covers in this concept
- iteration
- constraints
A prompt is a spec. Name the role, the audience, the task, the constraints, and the output shape. When it fails, change one thing: examples, delimiters, or the order of evidence. Keep prompts in git with a small eval set so a wording change is measurable, not tribal knowledge in a dashboard box. Score against failures from support logs, not only lab examples that already work.
Updated 2026-08-09 · Full learning path