Once a team has a prompt that works, it needs to store it, reuse it, improve it safely and know which version is running in production. Amazon Bedrock Prompt Management does that: it stores reusable prompts with variables, lets you test variants side by side, and saves numbered versions you can deploy and roll back. This lesson covers how it works and when the exam expects you to choose it.
Prompts and variables
In Prompt Management, a prompt is a saved, reusable template. You write the prompt text once and mark the parts that change as variables, such as product_name or customer_question. You supply values for those variables when you test the prompt, or at runtime when your application calls the model. Each saved prompt also stores the model or inference profile it should run on and its inference settings, such as temperature and maximum length. The whole configuration travels together rather than being scattered through application code.
Variants and testing
A prompt variant is an alternative configuration of the same prompt: different wording, a different model, or different inference settings. In the console's prompt builder you fill in test values for the variables, run the variants and compare their outputs side by side, then keep the one that works best. Bedrock's prompt optimization feature fits into this workflow too. It can generate a rewritten version of your prompt for a chosen model, shown next to the original so you can compare them.
Drafts and versions
When you save a prompt, you're editing its draft. You can keep changing and saving the draft as you iterate. When you're happy with a configuration, you create a version: a numbered snapshot of the draft at that moment. Versions are numbered upward from 1. The draft can keep changing while each version stays as it was when you created it.
Versions are what make prompt changes safe in production. Your application can use a specific version, so edits to the draft don't change live behavior until you choose to switch. If a new version performs badly, you point the application back to the previous one. The console can also compare two versions, highlighting what differs between their configurations and letting you run both against the same test values.
Using prompts in applications
There are two ways to put a managed prompt to work. You can specify the prompt when you run model inference, passing values for its variables. Or you can add it as a prompt node in an Amazon Bedrock Flows workflow. Either way, the prompt lives in one managed place instead of being copied into every application that uses it.
| Need | Prompt Management feature |
|---|---|
| Reuse one prompt for many inputs | Variables |
| Try different wording or models | Variants, compared side by side |
| Keep iterating without affecting production | The draft |
| Freeze a tested configuration | A version |
| Undo a bad change | Point back to an earlier version |
On the exam
Look for requirements about storing, reusing, versioning or comparing prompts across a team or applications. "Track changes to prompts and roll back" means versions. "Compare different wordings or models" means variants. Don't confuse Prompt Management with Bedrock Guardrails, which filters content, or Bedrock Knowledge Bases, which retrieves documents. Prompt Management manages the prompts themselves.