1.2.2. Where AWS Services Fit in the Lifecycle

💡 First Principle: AWS services map to lifecycle stages, not to arbitrary categories. When you know which stage you're in, you can immediately narrow down the relevant services—and the exam rewards this filtering ability.

The most important insight for exam preparation: SageMaker spans all four stages. It's not a single service—it's an ecosystem with components in every lifecycle stage. The exam frequently tests whether you know which SageMaker component handles a specific task. Data Wrangler is Stage 1. Training jobs and AMT are Stage 2. Endpoints and Pipelines are Stage 3. Model Monitor and Clarify are Stage 4. Canvas is the odd one out: it spans Stages 1 and 2 on its own, because it prepares the data and builds the model without code — which is why scenarios describing a team with no ML staff so often resolve to it.

Services outside SageMaker fill gaps where SageMaker isn't the best fit. AWS Glue handles large-scale ETL that Data Wrangler can't (think petabyte-scale data lakes). Amazon Bedrock provides access to foundation models when you don't want to train from scratch. CloudWatch monitors infrastructure metrics that Model Monitor doesn't cover (CPU utilization, endpoint latency). IAM and KMS secure the entire pipeline.

⚠️ Exam Trap: Don't confuse "SageMaker" the brand with a specific capability. A question asking about "bias detection" could mean SageMaker Clarify (Stage 2 — pre-training bias) or SageMaker Clarify (Stage 4 — post-deployment bias). Read carefully whether the question is about training data or production predictions.

Reflection Question: A company wants to detect if their sentiment analysis model's prediction distribution has shifted since deployment. Which lifecycle stage and which specific AWS service addresses this?