AWS generative AI services: AIF-C01 study notes and practice questions
What to know about AWS generative AI services for the AIF-C01 exam, from the Fundamentals of Generative AI domain (24% of the exam guide), then 5 free questions to test it. Part of the AIF-C01 study guide.
These questions ask which AWS service fits a generative AI need, and why use AWS for it.
The main services
Service
What it is
Amazon Bedrock
A fully managed service with API access to foundation models from Amazon and other providers; no servers to manage
Bedrock features
Knowledge Bases (RAG), Agents, Guardrails, model evaluation, fine-tuning and distillation, Prompt Management, Flows
Bedrock PartyRock
A free, no-code playground for trying generative AI apps
Amazon Bedrock AgentCore
Services to deploy and run AI agents securely at scale
Amazon SageMaker JumpStart
Deploy and fine-tune open models on infrastructure you choose
Amazon SageMaker AI
Build, train and host your own models
Amazon Q Business
An assistant that answers questions from company data
Amazon Q Developer
An AI assistant for software development and AWS tasks
Amazon Quick
AI agents, chat, research and dashboards over business data (it grew out of Amazon QuickSight)
Kiro
An agentic coding tool that turns requests into specs, designs and tasks
Amazon Nova
Amazon’s own family of foundation models, available in Bedrock
Why build on AWS
Security and privacy (your data isn’t used to train the base models in Bedrock), compliance, a choice of models, pay-as-you-go pricing, integration with other AWS services, and lower barriers to entry.
Cost factors
Option
Pricing
Fits
On-demand
Per input and output token
Testing, variable traffic
Batch inference
Lower per-token price for large offline jobs
Processing many prompts without waiting
Provisioned Throughput
Reserved model capacity per hour, with a term
Steady high volume; required for some custom models
Prompt caching
Reused prompt prefixes cost less and return faster
Long, repeated system prompts or documents
Larger models, longer prompts and longer answers all cost more.
Common traps
SageMaker AI for “no infrastructure to manage”. Bedrock is the managed, serverless choice.
Q Business vs Q Developer. One helps employees with company knowledge; the other helps developers.
Provisioned Throughput for spiky testing. On-demand fits variable use.
A logistics company is moving its AI agents into production on Amazon Bedrock AgentCore.
The platform team must assign each requirement to the AgentCore service that meets it. Each option is used at most once, and one option is not needed.
Match each item to one answer. An answer can be used more than once, and some may not be used.
Items
Host the agent code in a secure, serverless runtime that scales with demand
Keep key facts about each user so agents can recall them in later sessions
Expose existing APIs and Lambda functions to agents as MCP-compatible tools
Let agents call third-party services on behalf of users, using managed credentials
Trace, debug and monitor agent behavior in production
Answers
AA. AgentCore Observability, which traces, debugs and monitors agent performance.
BB. AgentCore Runtime, which hosts and scales agents in a secure, serverless environment.
CC. AgentCore Identity, which manages agent identities, access and authentication.
DD. AgentCore Browser, which gives agents a cloud browser to navigate sites and fill in forms.
EE. AgentCore Memory, which keeps context about users so agents can recall it later.
FF. AgentCore Gateway, which turns existing APIs and Lambda functions into MCP tools.
Show the answer and explanation
Correct matches
Host the agent code in a secure, serverless runtime that scales with demand: B. AgentCore Runtime, which hosts and scales agents in a secure, serverless environment.
Keep key facts about each user so agents can recall them in later sessions: E. AgentCore Memory, which keeps context about users so agents can recall it later.
Expose existing APIs and Lambda functions to agents as MCP-compatible tools: F. AgentCore Gateway, which turns existing APIs and Lambda functions into MCP tools.
Let agents call third-party services on behalf of users, using managed credentials: C. AgentCore Identity, which manages agent identities, access and authentication.
Trace, debug and monitor agent behavior in production: A. AgentCore Observability, which traces, debugs and monitors agent performance.
About each answer
A. Observability gives a unified view to trace, debug and monitor agent performance in production. It matches the need to watch agent behavior once it is live.
B. Runtime is a secure, serverless environment for deploying and scaling agents and tools. It matches the hosting need.
C. Identity manages agent identities, access and authentication, including the credentials agents need for outside services. It matches the third-party access need.
D. Browser lets agents use web applications by navigating sites and filling in forms. None of the five requirements is about web browsing, so this is the option that is not used.
E. Memory lets agents keep context across sessions, including long-term facts about each user. It matches the need to recall user facts in later sessions.
F. Gateway converts APIs, Lambda functions and existing services into Model Context Protocol tools that agents can call. It matches the need to expose existing services to agents.
Explanation
Each AgentCore service addresses one production need. Runtime hosts the agent, Memory keeps user context across sessions, Gateway exposes existing APIs as MCP tools, Identity handles credentials for outside services, and Observability traces and monitors behavior. Browser is for web navigation, which no requirement here asks for. This question tests matching each production need to its AgentCore service.
A start-up's developers want to build an AI agent in a few lines of Python.
The foundation model should decide the plan and which tools to call, rather than follow steps the team hand-codes, and the agent must use tools served over the Model Context Protocol (MCP).
Which option fits BEST?
Choose one answer.
AA. AWS Step Functions, with each agent step coded as an explicit state
BB. Amazon Bedrock AgentCore, a managed platform that hosts and operates agent code
CC. Strands Agents, an open-source SDK where the model picks tools, with MCP support
DD. Amazon Bedrock Flows, a visual builder that wires prompts and models into fixed steps
Show the answer and explanation
Answer: C
A. Wrong: Step Functions orchestrates workflows that the developer defines state by state. That is the hand-coded control the team wants to avoid, although the service suits fixed pipelines well.
D. Wrong: Flows lets teams design a fixed sequence of prompts, models and conditions. The team wants the model to plan its own steps, so a fixed flow misses that requirement.
Explanation
Strands Agents is an open-source SDK, first released by AWS, that puts the foundation model at the center of the agent. It supports MCP natively and works with models on Amazon Bedrock and other providers. AgentCore hosts and operates agents built with frameworks such as Strands, but it does not replace the SDK.
B. Wrong: On-demand Bedrock gives simple API access with no servers to manage, so the firm cannot pick the instance type or control the endpoint. It suits teams that do not need that control.
C. Wrong: Comprehend is a pre-trained language service for tasks such as entity and sentiment detection. It does not host a foundation model of the firm's choice, so it misses the control and fine-tuning needs.
D. Wrong: Rekognition Custom Labels trains models for image tasks. It cannot host or fine-tune an open-weight language model, so it does not meet the firm's needs.
Explanation
There are two broad routes to a foundation model on AWS. Amazon Bedrock is a managed API with no servers to choose. SageMaker JumpStart deploys open-weight models to endpoints you configure, which gives more control over instances and fine-tuning, but the customer runs more of the stack. Check each model's license before use.
A retail company's managers want to ask plain-language questions about sales data and internal documents, get answers with charts, and set up simple automated workflows.
They do not want to write code or manage models.
Which AWS service is designed for this?
Choose one answer.
AA. Amazon SageMaker AI, a platform for building and training custom ML models
BB. Amazon Quick, an AI-powered workspace with chat, dashboards and automation
CC. Amazon Inspector, which scans EC2 and container workloads for software vulnerabilities
DD. Amazon Transcribe, which converts speech in recordings into text
Show the answer and explanation
Answer: B
A. Wrong: SageMaker AI is built for data scientists who train and deploy their own models. It needs ML expertise and gives business managers no ready-made chat or dashboard experience.
B. Right: Correct. Amazon Quick is a fully managed service that combines natural language chat over connected data with dashboards and AI-powered workflows. It needs no infrastructure or ML expertise, and it evolved from Amazon QuickSight. Learn more: What is Amazon Quick? (opens in a new tab)
C. Wrong: Inspector scans EC2 instances, container images and other workloads for software vulnerabilities. It is a security tool, so it does not answer business questions or build workflows.
D. Wrong: Transcribe converts speech in audio and video into text. It does not analyze sales data, answer questions over documents or automate business tasks.
Explanation
Amazon Quick, which evolved from Amazon QuickSight, brings AI agents, chat over business data and documents, dashboards and automation to business users. Its flows automate repetitive tasks. Developers who need custom ML models use services such as SageMaker AI instead.
A development team wants an AI coding assistant that turns a feature request into requirements, a design and a task list that the team can review before any code is written.
Which AWS offering is built around this spec-driven approach?
Choose one answer.
AA. AWS CodeBuild, which compiles and tests existing code in managed build environments
BB. Amazon Bedrock Guardrails, which screen model inputs and outputs for unsafe content
CC. AWS CloudFormation, which provisions cloud infrastructure from declarative templates the team writes
DD. Kiro, an agentic coding service that drafts specs for team review
Show the answer and explanation
Answer: D
A. Wrong: CodeBuild runs build and test jobs for code that already exists. It does not turn a feature request into a specification or write the code.
B. Wrong: Guardrails screen model inputs and outputs for harmful or sensitive content. They do not plan or write software.
C. Wrong: CloudFormation provisions cloud resources from templates the team writes. It does not generate requirements, designs or code.
Kiro is an agentic coding service from AWS. In spec-driven development it turns a request into requirements, a design and implementation tasks. The developer reviews each phase before the tasks are run, so the plan is checked before any code is written.
Want more on AWS generative AI services? The 5 full-length practice sets have 18 more AWS generative AI services questions, 325 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →