AWS Certified AI Practitioner Roadmap

Choose, evaluate, and use AWS AI solutions responsibly. Follow four connected phases, reuse the same six selected core resources, and finish each phase with a practical learning artifact — not just completed videos. The optional YouTube videos supplement these resources; they do not replace any course or practice exam.

Exam at a glance: 100 USD per attempt, plus applicable taxes · 90 minutes · 65 questions · online or test center. Full official details, languages, scoring, and accommodations appear in Planning guidance. Verified October 20, 2026.

Phase 1 — Match Business Problems to AI Solutions (Weeks 1–2)

Start with the cloud and AI vocabulary needed to understand exam scenarios. This phase turns unfamiliar service names into clear business choices and covers the foundations behind Domain 1: Fundamentals of AI and ML, worth 20% of scored content.

Step 1 — Explain the AWS foundation behind an AI application

Learn what cloud services do before memorizing AI products. Focus on Amazon S3 for storage, Amazon EC2 for compute, AWS Lambda for serverless functions, IAM for access control, Regions and Availability Zones, the shared responsibility model, and consumption-based pricing. Recognize Amazon VPC as a networking boundary and the broad roles of Amazon Bedrock and Amazon SageMaker AI; configuring infrastructure is not required.

Practice outcome: Produce a one-page diagram showing where an application's data lives, where processing happens, who can access it, and which responsibilities belong to AWS versus the customer.

• AWS Cloud Practitioner Essentials — Free, beginner-friendly digital course from AWS. Study the relevant compute, storage, security, and pricing sections rather than completing the entire cloud curriculum.

Step 2 — Choose the right AI approach and AWS service

Distinguish AI, machine learning, deep learning, neural networks, and generative AI. Compare supervised, unsupervised, and reinforcement learning; then connect classification, regression, and clustering to business problems. Recognize structured and unstructured data, including tabular, time-series, text, and images. Understand training versus inference, labeled versus unlabeled data, overfitting, bias and fairness, and the basic ML lifecycle. Distinguish batch, real-time, asynchronous, and serverless inference at a use-case level.

Learn when accuracy is misleading and when precision, recall, or F1 matters more. Match speech-to-text, translation, sentiment analysis, text-to-speech, and conversational interfaces to Amazon Transcribe, Translate, Comprehend, Polly, and Lex. Include cases where ordinary rules or software are better than AI.

Trace the lifecycle from business objective and data preparation through training, evaluation, deployment, monitoring, and retraining. Recognize MLOps as repeatable operations across that lifecycle, and connect technical metrics to business outcomes such as cost per user and return on investment. This is conceptual knowledge, not a pipeline-building assignment.

Practice outcome: Build a ten-row decision table: business problem → AI technique → AWS service → success metric → reason not to use an alternative.

• Stephane Maarek: Ultimate AWS Certified AI Practitioner AIF-C01 — Paid video course with explanations, demonstrations, and practice material. Use its AI/ML foundations and AWS AI services sections as your structured teaching backbone.

Phase 1 project

Write an AI opportunity brief for a fictional online shop. Assess ticket classification, demand forecasting, and FAQ support. For each, choose AI, generative AI, or a non-AI approach; identify the AWS service, required data, and business metric. Combine the brief with your cloud diagram and decision table.

Phase 2 — Design and Evaluate Generative AI Solutions (Weeks 3–5)

Now move from recognizing services to explaining how a generative AI solution works. This phase gets the most time because Domain 2, Fundamentals of GenAI, is worth 24%, and Domain 3, Applications of Foundation Models, is worth 28%.

Step 3 — Select a foundation model and customization approach

Understand tokens, embeddings, vectors, context windows, multimodal models, and the foundation-model lifecycle. Compare models using task quality, supported languages, modality, latency, context length, and token-based cost — not popularity alone.

Explain when to use prompting, retrieval-augmented generation (RAG), or fine-tuning. RAG supplies retrieved information at inference time; fine-tuning changes model behavior through additional training. Neither automatically eliminates hallucinations. Also recognize pre-training, continued pre-training, instruction tuning, model distillation, and reinforcement learning from human feedback (RLHF) at a purpose-and-cost-tradeoff level — not an implementation level.

Recognize the purpose of Amazon Bedrock, SageMaker AI, and SageMaker JumpStart. At a conceptual level, learn agents, tools, memory, orchestration, and Model Context Protocol (MCP), plus the roles of Amazon Bedrock AgentCore and Strands Agents. Recognize Amazon Quick and Kiro where they appear in the current exam guide; implementation is not required.

Practice outcome: Create a comparison sheet for three use cases — document Q&A, brand-style writing, and an assistant that takes actions — and justify the model-selection criteria and customization approach for each.

• AWS Artificial Intelligence Practitioner Learning Plan — Official foundational learning plan. Focus on generative AI concepts and AWS capabilities to anchor your terminology in AWS's own explanations.
• Supplemental video (optional): AWS Power Hour: AI Practitioner | S2 E2 | Fundamentals of Generative AI (AWS Events) — Reinforce Domain 2 with explanations of generative AI, business use cases, capabilities, limitations, and AWS technologies. Revisit its prompting and Bedrock segments in Step 4.

Step 4 — Improve prompts and evaluate whether the application works

Practice clear instructions, relevant context, output constraints, zero-shot and few-shot examples, and reusable prompt templates. Understand how temperature and output length affect responses. Distinguish prompt engineering from context engineering: the latter also manages retrieved information, conversation history, tool results, and memory supplied to the model. Learn prompt versioning with Amazon Bedrock Prompt Management, recognize prompt injection and jailbreaking risks, and explain prompt caching as a model-selection and cost consideration.

Trace a simple RAG flow: documents → chunks → embeddings → vector search → retrieved context → answer. Understand the roles of Amazon Bedrock Knowledge Bases and vector stores; recognize Amazon OpenSearch Service and Amazon Aurora as examples of services that can store embeddings. Compare RAG with an agent that can call tools, and include document access permissions in your architecture sketch.

Evaluate outputs against a small reference set. Recognize human evaluation, ROUGE, BLEU, BERTScore, and LLM-as-a-judge at a purpose-and-limitations level; connect these to business metrics such as task completion, user satisfaction, and cost per interaction.

Practice outcome: Prepare ten FAQ test questions, three prompt versions, and a scorecard for correctness, grounding, completeness, and unsafe answers. Compare outputs in a permitted model playground or use a paper-based walkthrough if you do not want cloud charges.

• Stephane Maarek: Ultimate AWS Certified AI Practitioner AIF-C01 — Reuse the foundation-model, prompt engineering, RAG, customization, and evaluation sections. One consistent course helps you connect concepts rather than restart with another instructor.
• Supplemental video (optional): AWS Power Hour: AI Practitioner | S2 E3 | Applications of Foundation Models (AWS Events) — Official AWS session for Domain 3, covering foundation-model application design, prompt engineering, RAG, agents, customization, and evaluation. Use it to reinforce this phase, then check the current exam guide for newer topics.

Phase 2 project

Design and evaluate a shop FAQ assistant. Use a small set of fictional policy documents. Deliver an architecture sketch, a model-selection rationale, a prompt template, a RAG-versus-fine-tuning decision, and your ten-question evaluation scorecard. Include questions the assistant should decline or escalate because the documents do not contain an answer. A live deployment is optional, not an exam prerequisite.

Phase 3 — Make AI Responsible, Secure, and Governed (Weeks 6–7)

Once you can explain the application, learn how to judge whether it is safe to use. This phase covers Domain 4, Guidelines for Responsible AI, and Domain 5, Security, Compliance, and Governance for AI Solutions — 14% each.

Step 5 — Identify AI harms and choose safeguards

Learn fairness, bias, inclusivity, robustness, safety, transparency, and explainability. Recognize how unrepresentative data, misleading outputs, intellectual-property issues, and overreliance on AI can harm users.

Understand where Amazon Bedrock Guardrails, model evaluations, model cards, human review, and subgroup testing help — and what they cannot guarantee. Distinguish explaining a model's behavior from simply showing its output.

Practice outcome: Create a five-risk register for your FAQ assistant. Give each risk a concrete example, a safeguard, a test, and a condition requiring human escalation.

• AWS Exam Prep Plan: AWS Certified AI Practitioner — Use the Domain 4 review and practice activities, not the entire plan. Official scenario questions help distinguish responsible-AI principles that sound similar; some activities require a subscription.

Step 6 — Explain the security and governance controls a scenario needs

Connect least-privilege IAM access, encryption, private connectivity, data protection, and logging to specific risks. Distinguish Amazon Macie for sensitive-data discovery, AWS CloudTrail for API activity, AWS Config for resource configuration, AWS Artifact for compliance reports, and Amazon Inspector for vulnerability management.

Cover data lineage, retention, residency, ownership, review responsibilities, and auditability. Understand that guardrails do not replace IAM, RAG is not an access-control system, and using an AWS service does not automatically make an application compliant.

Practice outcome: Produce a control matrix: risk → control → AWS service or process → owner → evidence. Include sensitive-data exposure, unauthorized access, prompt injection, and missing audit records.

• AWS Exam Prep Plan: AWS Certified AI Practitioner — Reuse its Domain 5 review and practice activities to learn service selection and governance decisions at the exam's conceptual depth.

Phase 3 project

Conduct a launch review of your FAQ assistant. Combine the risk register, control matrix, evaluation findings, and human-escalation rules into a two-page go/no-go brief. State what must change before launch and what should continue to be monitored afterward. You are learning to identify controls, not implement an enterprise governance framework.

Phase 4 — Demonstrate Exam Readiness (Week 8)

Convert knowledge into reliable scenario reasoning. The goal is to explain your choices on unfamiliar questions, not recognize answers you have memorized.

Step 7 — Diagnose and repair weak areas

Take an unseen timed practice exam before opening explanations. Review every incorrect or uncertain answer, including lucky guesses. Record the missing concept, why your chosen answer was wrong, and what clue makes the correct answer fit. Return only to the relevant course section or official guide topic.

Before your first full mock, use the official practice question set or pretest linked from the AWS exam page to check question style and gaps. Keep short revision notes in your own words; compare them with the current guide rather than relying on someone else's notes or an older course's service list.

Practice outcome: Create a domain-by-domain error log and explain ten previously missed scenarios without looking at the answer choices.

• Tutorials Dojo: AWS Certified AI Practitioner AIF-C01 Practice Exams — Paid question bank with timed, review, and section-based modes plus explanations. Particularly useful for targeted remediation after a first honest attempt.
• Supplemental video (optional): AWS Power Hour: AI Practitioner | S2 E4 | Security and Responsible AI (AWS Events) — Official AWS session reviewing Domains 4 and 5; use it to reinforce responsible-AI and security concepts, then check the current exam guide for newer topics.

Step 8 — Validate readiness under exam conditions

Take the official practice exam in one uninterrupted sitting, using the standard 90-minute limit unless you have an approved accommodation. Practice multiple-choice, multiple-response, ordering, and matching questions; older banks may not cover every current format. Review your reasoning afterward, then check your weakest topics against the current AIF-C01 exam guide, especially agentic AI concepts and recently named services. Answer every exam question: unanswered questions are incorrect, and AWS does not penalize guessing.

As a personal readiness rule — not an AWS requirement — aim for at least 80% on two different unseen, full-length practice tests and be able to explain the alternatives. Repeated scores on the same questions do not establish readiness, and raw practice percentages do not map directly to the exam's scaled score.

Practice outcome: Finish with two first-attempt practice results, a resolved error log, and a one-page summary covering model selection, RAG versus fine-tuning, responsible AI, security, and governance.

• AWS Official Practice Exam: AI Practitioner AIF-C01 — Subscription-based official assessment. Best used near the end to check your understanding against AWS-authored questions rather than as an early memorization exercise.

Phase 4 project

Complete your exam-readiness portfolio. Assemble the opportunity brief, solution design, evaluation scorecard, launch review, and practice-test evidence. Give a five-minute explanation of your solution without notes, including one reason to use AI, one reason not to, and its biggest remaining risk. Extend revision if you cannot yet explain these decisions comfortably.

Recap

Planning guidance

Use a repeatable weekly rhythm. Spend roughly half your study time on instruction, a quarter on the phase artifact, and a quarter on retrieval practice and reviewing mistakes. This is a suggested pace, not a guaranteed preparation time. Complete beginners who need the full cloud prerequisite should allow extra time.

Skip what you already know. If you can explain S3, EC2, Lambda, IAM, shared responsibility, and AWS pricing confidently, skip Step 1 and put that time into foundation models or weak domains. You do not need to earn Cloud Practitioner first.

Keep the scope foundational. AWS says coding models, implementing pipelines, performing hyperparameter tuning, and deploying infrastructure are outside this exam's target scope. Projects here are learning aids, not requirements for certification or proof of ML engineering readiness.

Control costs. The selected resources mix free and paid content; some AWS prep activities and the official practice exam require a subscription. The prep plan may already include activities linked separately here, so check access before buying overlapping material. A personal AWS account and live deployment are not required for the paper-based projects. If you use live services, check model and Region availability, review pricing, set budget alerts, and clean up resources; alerts are not a hard spending cap.

Use the comparison roadmaps selectively. The supplied GitHub roadmap reinforces cloud foundations → AI/ML → GenAI/Bedrock → security, governance, and costs. This version retains its eight-week pace and scenario-based projects rather than adding the repository's broader EMR, Redshift, DataBrew, replication, or infrastructure labs as certification prerequisites. Those labs are optional cloud-career enrichment, not required preparation for this foundational exam.

The supplied Medium roadmap supports using one teaching backbone, revising notes, and then taking practice exams. Its suggested 85–90% practice results are an optional stretch target, not an AWS pass requirement. Course-gap observations are the author's experience, not a current syllabus guarantee. Use AWS — not either community guide — for exam rules; in particular, the current guide includes ordering and matching as well as multiple-choice and multiple-response questions.

Know the exam. The details below were checked against the official AWS exam overview, AIF-C01 exam guide, and AWS before-testing policies on October 20, 2026. Recheck the booking page before paying.

Before booking: Open Schedule your exam on the official AWS overview and use your AWS Certification account. Request any needed accommodation before registration, confirm exam code, language, final fee and taxes, and choose test-center or online delivery. Check identification, name-matching, check-in, and appointment-change rules in the provider's instructions. Reserve the exam when your readiness evidence supports it — not simply because Week 8 has arrived.

Mistakes to avoid

• Memorizing service names without understanding the problem each service solves.
• Assuming RAG, fine-tuning, guardrails, and agents are interchangeable — or that any one eliminates hallucinations.
• Ignoring responsible AI and governance because they seem less technical.
• Trusting an older course as the final syllabus instead of checking the current exam guide.
• Using exam dumps or treating repeat-attempt practice scores as evidence of understanding.
• Spending weeks on Python, model training, or production deployment instead of this foundational exam's decision-making skills.