The service surface, as a map

What you will be able to do

• Group the in-scope AWS services by the question each one answers
• Use the current name for every service the exam guide lists
• Explain why a service closed to new customers can still appear in a question
• Identify the near-miss services the blueprint puts out of scope
• State the depth ceiling the exam sets for every service

The exam guide publishes a list of in-scope AWS services, grouped by AWS's own service categories: Analytics, Compute, Database, Developer Tools, Machine Learning and so on. That grouping is fine for a product catalogue and useless for study, because it puts Amazon Quick next to Amazon Redshift and Kiro next to Amazon Q while separating Amazon Bedrock from the vector stores its knowledge bases run on.

A more useful arrangement is by the question a service answers. Exam items rarely ask what a service is. They describe a situation and ask which service fits, so the skill being tested is going from a problem to a name.

Eight questions, and what answers each#

"I want a foundation model behind an Application Programming Interface (API)." Amazon Bedrock is the managed access point to foundation models from several providers, including Amazon's own Amazon Nova family. Its features carry their own names and are examined individually: Bedrock Knowledge Bases for managed retrieval augmented generation, Bedrock Guardrails for input and output safeguards, Bedrock Prompt Management for versioned prompts and Bedrock Model Evaluation for comparing models.

"I want to build, train, tune or self-host a model." Amazon SageMaker Artificial Intelligence (AI). Amazon SageMaker JumpStart sits inside it and offers pretrained open source and third-party models, solution templates and example notebooks, which is the path when you want a model you can fine-tune and host yourself rather than call over an API.

"I want a ready-made model for one specific task." This is the oldest and most heavily examined group, and the items are usually straightforward matching:

One name is regularly filed with that group and does not belong in it. Amazon Augmented AI (A2I) returns no prediction of its own: it routes another model's low-confidence predictions, or a random sample of them, to a human workforce, so it sits after a prediction rather than producing one. AWS has closed it to new customers, and it stays on the exam's in-scope list, so it remains the expected answer when an item asks which service adds human review.

"I want an agent." Two names, and they are not alternatives to each other. Strands Agents is an open source SDK, licensed under Apache 2.0, that you write an agent in: a model, a system prompt and a set of tools, with the framework running the loop. Amazon Bedrock AgentCore is the managed platform an agent runs on, with modular components for hosting, memory, tool access, identity and observability, and it works with any framework and any model rather than only with Bedrock. Strands is what you build with. AgentCore is where it lives.

"I want to give a person an assistant." Amazon Q is AWS's assistant family, filed by the exam guide under Developer Tools. Amazon Quick is the AI-powered service for business users, covering natural-language chat over connected data, task automation, research and app building, with Amazon Quick Sight as its data visualisation and business intelligence feature. Kiro is the agentic coding service, built on Amazon Bedrock, that turns a prompt into specifications before it writes code. AWS Transform applies agents to migration and modernisation work: mainframes, VMware estates and .NET applications.

"Where does the data live?" Amazon S3 and S3 Glacier for objects. AWS Glue, Glue DataBrew, AWS Lake Formation, Amazon EMR, Amazon Redshift and AWS Data Exchange for preparing and moving it. For embeddings, the objectives name four stores: Amazon OpenSearch Service, Amazon Aurora, Amazon Neptune and Amazon RDS for PostgreSQL. Amazon DynamoDB, Amazon DocumentDB and Amazon ElastiCache are in scope as databases without being the answer to a vector question.

"How do I run it and what does it cost?" Amazon EC2 and AWS Lambda for compute, Amazon ECS and Amazon EKS for containers, Amazon CloudFront and Amazon VPC for delivery and networking, AWS Budgets and AWS Cost Explorer for the bill.

"How do I secure and govern it?" AWS IAM, AWS KMS, Amazon Macie, AWS Secrets Manager and Amazon Inspector on the security side. AWS CloudTrail, Amazon CloudWatch, AWS Config, AWS Artifact, AWS Trusted Advisor and the AWS Well-Architected Tool on the governance side. Domain 5 also names AWS PrivateLink, Bedrock Guardrails, AgentCore Identity and Policy in AgentCore as controls rather than as products in their own right.

The names that changed#

This is where older study material fails, and it fails confidently. A run of renames and status changes has landed since 2024, and several study guides, video courses and question banks written before 2026 still use the superseded names.

Bedrock Agents Classic deserves its own note because the conflation is so easy. Amazon Bedrock itself is unaffected, as are Bedrock Knowledge Bases and Bedrock Guardrails. It is specifically the 2023 agent orchestration feature that went into maintenance mode. The exam guide lists AgentCore in its in-scope services and names AgentCore components in domain 5, and it does not list Bedrock Agents at all, which tells you where the exam's attention has moved.

Closed to new customers is not deleted#

It is tempting to treat everything in the table above as gone. That is the wrong reading, and it will lose you items.

Amazon Q Business still runs for the accounts that already had it. Bedrock Agents Classic still runs for accounts with prior activity. Both still have documentation. So both remain fair game for a question, and the safest way to hold the knowledge is as a pair: what the service does, and what you would choose instead for something new. If an item asks which service a company should adopt today for a natural-language assistant over its own documents, the answer is Amazon Quick. If an item describes a company that already has Amazon Q Business, the answer is about Amazon Q Business.

What the blueprint puts out of scope#

The guide publishes an out-of-scope service list as well, and the entries worth knowing are the near misses inside categories that are otherwise in scope: AWS Clean Rooms, Amazon MSK, Amazon Cognito, Amazon GuardDuty, AWS Security Hub, AWS WAF, AWS Control Tower, AWS Organizations, AWS IAM Identity Center, AWS X-Ray, AWS HealthOmics and AWS DeepComposer are all out. Amazon MemoryDB was removed from the in-scope list in the April 2026 revision. Both lists are non-exhaustive by AWS's own statement, so a service on neither list, Amazon Kendra being the obvious example, is unlisted rather than banned.

More important than the service list is the job task list, because it sets the depth ceiling for everything you study. The guide states plainly that the target candidate is not expected to develop or code models and algorithms, do data or feature engineering, tune hyperparameters, build pipelines or infrastructure, perform mathematical or statistical analysis of models, implement security and compliance protocols, or develop governance frameworks.

Read that list as a permission slip. You need to know what hyperparameter tuning is, why it matters and roughly where it sits in the lifecycle. You do not need to know how to run it. Every time a topic in this guide starts to feel like it is demanding implementation detail, that list is the reason you can stop.

The next lesson covers what changed in the April 2026 revision, which is why several of the names in this one would have been wrong a year ago.