Every AWS service on the AIF-C01 exam β and why it's there
A practical breakdown of the 21 AWS services tested on the AI Practitioner exam: what each one does, which exam domain it maps to, and how deep you need to go.
The AIF-C01 exam covers a surprisingly wide surface of AWS services. Some of them β Bedrock, SageMaker β are the obvious headliners. Others β Macie, KMS, CloudWatch β show up in scenario questions where the answer hinges on knowing the right tool for a security or governance requirement. If you walk in knowing only the generative AI services, you'll leave points on the table.
This guide maps every AWS service that appears on the AIF-C01 exam, explains what it does in plain English, and tells you which exam domain it belongs to. I've grouped them the way the exam thinks about them: core generative-AI services first, then specialized pre-built AI services, then governance and security infrastructure.
Core generative-AI services
These are the services you'll spend the most study time on. They appear across Domains 2, 3, and 4 β the bulk of the exam.
Amazon Bedrock
Bedrock is the centerpiece of the AIF-C01 exam. It's a fully managed service that gives you API access to foundation models from Anthropic (Claude), Meta (Llama), Mistral, AI21, Cohere, Stability AI, and Amazon's own Titan family. You don't manage infrastructure, you don't train anything β you send prompts and get responses.
Exam angle: Domain 3 (Applications of Foundation Models) leans hard on Bedrock. Expect scenario questions on model selection (when to pick Claude vs. Titan vs. Llama), inference parameters (temperature, top-p, max tokens), and when Bedrock is the right choice vs. self-hosting on SageMaker.
Amazon Bedrock Knowledge Bases
This is AWS's managed RAG (retrieval-augmented generation) solution. You point it at S3 documents, it chunks them, generates embeddings, stores them in a vector database, and lets a foundation model retrieve relevant passages before generating a response. It handles citation tracking automatically.
Exam angle: RAG is the canonical answer to "how do you ground a foundation model in your private data without fine-tuning?" You'll see this scenario multiple times on the exam. Know the difference between RAG and fine-tuning β the exam tests it explicitly.
Amazon Bedrock Guardrails
Guardrails is a policy layer that sits between the user and the model. It can filter harmful content, block specific topics you define, redact PII from inputs and outputs, and ground responses to reduce hallucinations. You configure it declaratively β no custom code required.
Exam angle: Domain 4 (Guidelines for Responsible AI) asks how to prevent a model from generating unsafe or hallucinated content. Guardrails is the named AWS answer. Know the four types of protections it offers: content filters, denied topics, PII redaction, and grounding checks.
Amazon Bedrock Agents
Agents give a foundation model the ability to take actions β call APIs, query knowledge bases, execute multi-step workflows. You define "action groups" that describe what tools the model can use, and Agents handles the orchestration loop (think β act β observe β respond).
Exam angle: Domain 3 includes "AI orchestration" scenarios. The exam distinguishes Agents (multi-step, tool-using) from plain Bedrock inference (single prompt β single response). If a question describes a chatbot that needs to look up order status, modify a database, and respond β the answer is Agents.
Amazon Q
Amazon Q is AWS's packaged AI assistant product. Q Developer lives inside IDEs and the AWS Console to help with coding, debugging, and infrastructure questions. Q Business connects to enterprise data sources (SharePoint, Confluence, Salesforce) for internal Q&A.
Exam angle: AIF-C01 introduces Q as the "use this if you don't want to build anything" option. Questions test when to pick Q (pre-built, opinionated) vs. Bedrock (customizable, build-your-own).
Amazon SageMaker
SageMaker is the full ML platform β notebooks, training jobs, hyperparameter tuning, managed inference endpoints, and MLOps pipelines. It's the service you use when you're building and training custom models, not just calling pre-trained ones.
Exam angle: Domain 2 (Fundamentals of AI and ML) references SageMaker for the complete ML lifecycle. The exam tests whether you know the difference between training (SageMaker), inference hosting (SageMaker endpoints), and API-based model access (Bedrock).
Amazon SageMaker JumpStart
JumpStart is a model catalog inside SageMaker. It offers pre-trained foundation models and task-specific models with one-click deployment and fine-tuning notebooks. Think of it as an App Store for ML models that you can deploy into your own AWS account.
Exam angle: The exam distinguishes three ways to use a model: (1) call a hosted API through Bedrock, (2) deploy a pre-trained model from JumpStart into your own endpoint, (3) train from scratch on SageMaker. JumpStart is option 2.
Amazon SageMaker Clarify
Clarify is a bias detection and model explainability tool. It computes SHAP feature attributions (which input features drove this prediction?), measures pre-training and post-training bias across demographic groups, and works on both tabular models and foundation model outputs.
Exam angle: Domain 4 (Responsible AI) tests bias detection and explainability. When a question asks "how do you detect if your model is biased against a demographic group?" β Clarify is the answer.
Specialized pre-built AI services
These services solve specific AI tasks without requiring you to train or manage any models. The exam tests your ability to pick the right one for a given use case.
Amazon Comprehend
Managed NLP: sentiment analysis, entity recognition, key-phrase extraction, language detection, and PII identification. No training needed β send text, get structured results.
Exam angle: "I need to analyze customer feedback for sentiment" β Comprehend. "I need to find PII in a document corpus" β Comprehend (or Macie for S3 scanning). Know the boundaries.
Amazon Rekognition
Computer vision: label detection, face analysis, content moderation, celebrity recognition, video activity detection. Image or video in, structured labels out.
Exam angle: "I have images, no ML team" β Rekognition. The exam also uses Rekognition's content moderation feature in responsible-AI scenarios.
Amazon Textract
Document understanding: extracts text, key-value pairs, tables, and form fields from PDFs and scanned images. Goes beyond basic OCR β it understands document structure.
Exam angle: "Extract data from invoices/receipts/forms" β Textract. The exam specifically tests the difference between Textract (structured document extraction) and Rekognition DetectText (generic OCR on photos).
Amazon Transcribe
Speech-to-text with speaker identification, custom vocabularies, real-time streaming, and specialized variants for medical and call-center audio.
Exam angle: Pairs with Comprehend in pipeline questions: "transcribe a call center recording, then analyze sentiment" β Transcribe + Comprehend.
Amazon Polly
Text-to-speech with neural and generative voices across dozens of languages. Supports SSML for fine control over pronunciation, pacing, and emphasis.
Exam angle: "Convert text to speech" β Polly. The exam tests when Polly (pre-built TTS) is sufficient vs. when you'd need a custom voice model on SageMaker.
Amazon Translate
Neural machine translation across 75+ languages, with custom terminology support for domain-specific phrasing.
Exam angle: Straightforward β "translate content into multiple languages without training a model" β Translate.
Amazon Lex
Conversational AI for building chatbots with intents, slots, and speech output via Polly. Powers Alexa-style interactions.
Exam angle: Lex is the pre-built chatbot framework. The exam contrasts it with Bedrock Agents β Lex for structured, intent-based conversations; Agents for open-ended, tool-using AI assistants.
Amazon Kendra
ML-powered enterprise search across documents, SharePoint, Confluence, and databases. Understands natural-language queries, not just keyword matching.
Exam angle: Kendra appears as a retrieval-only alternative to Bedrock Knowledge Bases. When the question says "search, no generation needed" β Kendra. When it says "search and synthesize an answer" β Knowledge Bases.
Security, governance, and supporting infrastructure
These services aren't AI-specific, but they show up in Domain 5 (Security, Compliance, and Governance) questions. The exam tests whether you know how to secure and monitor AI workloads.
AWS IAM
Users, roles, policies, federation, and least-privilege permissions for every AWS service call. Not AI-specific, but the exam tests AI-specific IAM patterns.
Exam angle: "How do you restrict which models a developer can invoke in Bedrock?" β IAM policies. "How does a SageMaker notebook access training data in S3?" β IAM role. Least-privilege is the recurring theme.
AWS KMS
Managed cryptographic keys for encrypting data at rest β training datasets, model artifacts, inference logs.
Exam angle: "How do you encrypt model weights at rest with a customer-managed key?" β KMS. This comes up in compliance-focused questions.
Amazon Macie
Sensitive-data discovery service that uses ML to find PII, credentials, and financial data in S3 buckets.
Exam angle: "Before feeding documents to a RAG pipeline, how do you check for PII in the source data?" β Macie scans the S3 bucket; Guardrails filters at inference time. The exam tests both layers.
Amazon CloudWatch
Metrics, logs, and alarms across all AWS services. For AI workloads: Bedrock invocation logs, SageMaker endpoint latency, model-monitor drift alerts.
Exam angle: "How do you monitor model performance in production?" β CloudWatch metrics + alarms. "How do you audit which prompts were sent to Bedrock?" β CloudWatch Logs with Bedrock invocation logging enabled.
How to study this
Don't try to memorize every feature of every service. The exam tests pattern recognition: given a scenario, which service solves it? Here's the mental model:
- "I want to call a foundation model via API" β Bedrock
- "I want to ground a model in my private data" β Bedrock Knowledge Bases (RAG)
- "I want the model to take actions" β Bedrock Agents
- "I want safety rails on model output" β Bedrock Guardrails
- "I don't want to build anything, just use AI" β Amazon Q
- "I need to train a custom model" β SageMaker
- "I need a pre-trained model I can deploy myself" β SageMaker JumpStart
- "I need to check my model for bias" β SageMaker Clarify
- "I need to [analyze text / see images / extract documents / transcribe audio / synthesize speech / translate / build a chatbot / search documents]" β the corresponding specialized service
- "I need to secure / encrypt / scan / monitor" β IAM / KMS / Macie / CloudWatch
Run through practice questions with this framework and you'll find you can answer most service-selection questions in under 30 seconds.
Source: AWS AI Practitioner (AIF-C01) exam guide v1.1 (2024-08), AWS documentation as of May 2026.