Prompt Engineering

The quality of what a foundation model gives you depends less on the model than on how you ask. Prompt engineering is the practice of shaping that input so the model returns something useful, consistent, and safe. This topic takes a prompt apart, names the techniques that scale from simple to hard tasks, and covers what happens when the input turns hostile or the prompt has to change in production.

What This Topic Covers

• The anatomy of a prompt: instruction, context, input, and output indicator, plus prompt templates and the zero-shot versus few-shot distinction
• Prompting techniques along a ladder of scaffolding: zero-shot, one-shot, few-shot, chain-of-thought, and prompt chaining
• Best practices for clarity: specific instructions, instruction placement, separators, fallback answers, and negative prompts
• Prompt attacks and risks: prompt injection, jailbreaking, and prompt leaking, with layered defenses including Amazon Bedrock Guardrails
• Prompt versioning and management: treating prompts like code with Amazon Bedrock Prompt Management, drafts versus immutable versions, and safe rollback

Why It Matters

Prompt engineering sits inside the heaviest domain on the AWS Certified AI Practitioner exam, and its questions are practical. You will be shown a prompt and asked to name its parts, handed a task and asked which technique fits, or given an attack and asked which defense stops it. Knowing that temperature is not a security control, that a version is immutable while a draft is not, and that prompt injection is the customer's responsibility under the shared responsibility model is the kind of detail these questions turn on.

The skill outlasts the exam. Prompting is the cheapest, fastest way to change what a model does, no training, no infrastructure, just better input, and it is the first tool any team reaches for before spending on fine-tuning or new models.