Frequently Asked Questions
Do I need AWS ML experience for AIP-C01?
Yes. This is a professional-level exam. 2+ years of hands-on AWS AI/ML experience is strongly recommended. If you have not worked with SageMaker, Bedrock, or similar services in production, the implementation and troubleshooting domains will be difficult. AWS Certified Machine Learning Specialty or equivalent knowledge provides a solid foundation.
How many practice questions should I aim for?
Aim for at least 800 unique questions across all domains, ensuring you prioritize quality over quantity. ExamOS provides over 3,000 high quality practice questions covering all domains and designed to explain the rationale behind both correct and incorrect answers.
When should I book the exam?
Once you are consistently hitting 80% or above on Legend mode across two or more sessions, and you feel comfortable explaining GenAI architecture decisions for production scenarios, you are ready. Trust your preparation.
How long is the exam?
170 minutes for 75 questions (65 scored, 15 unscored). That averages to roughly 2 minutes and 16 seconds per question. Scenario-heavy questions with long context may take longer, so keep an eye on the clock. Answer every question; the unscored ones are indistinguishable from scored ones.
What is the passing score?
750 out of 1000. AWS uses scaled scoring. The raw number of correct answers needed varies slightly between exam versions. Do not try to calculate how many questions you can afford to miss.
Can I use reference materials during the exam?
No. The AIP-C01 exam is closed-book. You cannot access documentation, websites, or any external resources during the test. All questions are designed to be answerable from your knowledge of AWS GenAI services, architecture patterns, and best practices.
What is the most important domain to study?
Foundation Model Integration, Data Management, and Compliance (31%) is the heaviest domain. Combined with Implementation and Integration (26%), the first two domains represent 57% of the exam. RAG and agents are the core topics within these domains and appear in the majority of scenario questions.
What is the difference between Bedrock and SageMaker for GenAI?
Bedrock provides managed access to foundation models through an API. You use it for inference, RAG, agents, and guardrails without managing infrastructure. SageMaker is for training, fine-tuning, and deploying custom models with full control over the training process. The exam tests both and expects you to know when each is the right choice. Bedrock is for consuming foundation models. SageMaker is for building and customizing them.
What is the trade-off framework for fine-tuning vs RAG vs prompt engineering?
The exam tests this distinction frequently:
- Prompt engineering is the fastest and cheapest approach. Use it when the model's existing knowledge is sufficient and you need to guide output format or behavior.
- RAG is best when you need the model to access specific, up-to-date, or proprietary information that was not in its training data.
- Fine-tuning is best when you need the model to consistently follow a specific style, format, or domain-specific behavior that prompting alone cannot achieve reliably.
The binding constraint in the scenario (cost, accuracy, data freshness, latency) usually points to one of these three.
What are the most common ways people fail?
- Treating every AI question as a Bedrock question. Some scenarios require SageMaker for fine-tuning or custom training. Knowing when Bedrock is not enough is essential.
- Not understanding RAG depth. The exam goes beyond basic retrieve-and-generate. You need to understand chunking strategies, embedding model selection, hybrid search, query rewriting, and reranking.
- Ignoring AI safety. At 20% of the exam, safety and governance are not a footnote. Bedrock Guardrails, prompt injection defenses, and compliance frameworks show up regularly.
- Confusing on-demand and provisioned throughput. Know the cost and latency trade-offs and when each model is justified.
- Not practicing agent design. Bedrock Agents with action groups, function calling, and multi-agent orchestration are core topics. If you have not built one, you will struggle with these questions.
What are the three biggest preparation mistakes?
- Studying documentation without building. AIP-C01 is a practical exam. If you have not built a RAG pipeline, configured a Bedrock Guardrail, or deployed a fine-tuned model, you will struggle with scenario questions.
- Skipping safety and governance depth. Candidates with strong development backgrounds often focus on implementation and neglect Domain 3. At 20% of the exam, that is a costly oversight.
- Not practicing under time pressure. 75 questions in 170 minutes is generous, but scenario-heavy questions consume more time than you expect. Practice timed quizzes to build pacing.
Is there a renewal option?
Yes. AIP-C01 is valid for 3 years. To renew, you retake the current version of the exam. AWS also offers recertification pathways through their certification portal. Check the AWS Certification portal for the most current renewal options.
Quick-Reference: What Is on the Exam
For the full, always-current list, visit the official AIP-C01 exam guide.
| Domain |
Key Skills |
| Foundation Model Integration, Data Management, and Compliance (31%) |
Amazon Bedrock API (model invocation, streaming, response handling), model selection criteria (cost, latency, task suitability), prompt engineering (zero-shot, few-shot, chain-of-thought, system prompts), fine-tuning vs RAG vs prompt engineering selection, RAG architecture (chunking, embeddings, retrieval, generation), Knowledge Bases for Bedrock, vector stores (OpenSearch Serverless, pgvector), advanced RAG (query rewriting, reranking, hybrid search), embedding models (Titan, Cohere), data preparation and compliance requirements |
| Implementation and Integration (26%) |
Bedrock Agents (planning, orchestration, action groups, Lambda integration), function calling and tool schemas, multi-agent collaboration patterns, model fine-tuning (SageMaker JumpStart, LoRA, QLoRA, PEFT), model deployment (real-time, serverless, batch), integration patterns (API Gateway, Lambda, Step Functions, streaming), SageMaker Pipelines for GenAI workflows |
| AI Safety, Security, and Governance (20%) |
Bedrock Guardrails (content filters, denied topics, PII redaction), prompt injection (direct and indirect, mitigation strategies), encryption (KMS, VPC endpoints for Bedrock), auditing (CloudTrail, Config, AWS Artifact), responsible AI (bias detection, explainability, model cards), compliance (HIPAA, GDPR, FedRAMP for AI workloads) |
| Operational Efficiency and Optimization (12%) |
Cost management (token usage, prompt compression, model routing), latency optimization (streaming, caching, model distillation), provisioned throughput (on-demand vs reserved, cost trade-offs), SageMaker inference optimization, monitoring (CloudWatch metrics, custom dashboards, alarms), cost allocation (tagging, Cost Explorer, Budgets) |
| Testing, Validation, and Troubleshooting (11%) |
Evaluation metrics (ROUGE, BERTScore, groundedness, relevance), model evaluation (automatic vs human), A/B testing (comparing models, prompts, configurations), common failure modes and debugging, performance benchmarking (throughput, latency, cost per inference), model monitoring (data drift, prediction drift, quality drift) |