The same week, a municipal greenhouse needs a prompted foundation model in an app, a custom-trained wilt model, and a director dashboard of weekly detections. Which mapping is accurate?

Three municipal greenhouse jobs in one week map to three different AWS surfaces: a prompted foundation model in an application, a custom-trained wilt model, and a director dashboard of weekly detections. Amazon Bedrock is the path for consuming a prompted foundation model without standing up training infrastructure. Amazon SageMaker AI is the path for building and hosting the custom wilt model the greenhouse trains itself. Amazon Quick is the director-facing dashboard for weekly detection visibility. That mapping keeps each job on the service whose primary purpose matches the work.

Using Amazon Quick to host the custom wilt model and Bedrock to draw the dashboard fails because it swaps the training/hosting surface with the analytics surface and the prompted-FM surface with visualization. Using SageMaker AI to consume the prompted FM and Quick to train the wilt model fails for the same conceptual mix-up: prompted FM consumption is Bedrock’s job, and Quick does not train custom models. Inventing one service that does all three under a new name fails because the practitioner map is about recognizing existing service-to-job boundaries, not collapsing them into a fictional SKU.

Exam caveat: AIF-style questions often present multiple needs in one scenario and expect a clean Bedrock / SageMaker AI / Quick (or similar) split rather than a single catch-all product. Operational check: for each greenhouse deliverable, write the owning service beside it—prompted FM app → Bedrock, custom wilt training → SageMaker AI, weekly director view → Quick—and refuse any design that assigns those jobs to the wrong counter.