AWS Generative AI Developer Professional (AIP-C01) Study Guide

The AWS Certified Generative AI Developer – Professional exam (AIP-C01) tests whether you can put foundation models into real, production applications on AWS: retrieval, agents, guardrails, cost control and evaluation. It costs 300 USD, runs 180 minutes for 75 questions (65 of them scored), and you need 750 out of 1,000 to pass.

This guide walks through the five domains in AWS’s published exam guide, what each one covers, the services the guide leans on most, and a six-week study plan weighted to the blueprint. Every exam fact below comes from AWS’s own published pages; nothing here comes from the live exam.

AIP-C01 at a glance

Checked against AWS’s AIP-C01 certification page, the AIP-C01 exam guide and the AWS Recertification page on October 6, 2026.

Who AIP-C01 is for

AWS writes this exam for people already working as generative AI developers. The target candidate in the exam guide has two or more years building production-grade applications, some general AI/ML or data engineering background, and a year of hands-on generative AI work. AWS also expects working knowledge of AWS compute, storage and networking, security and identity, infrastructure as code, monitoring and observability, and cost optimization.

Just as useful is what the guide puts out of scope: model development and training, advanced ML techniques, and data and feature engineering. This is an integration and operations exam. You will not be asked to train a model from scratch; you will be asked how to wire a model into an application safely, cheaply and reliably.

No earlier certification is required. AWS says candidates could benefit from first earning the AWS Certified AI Practitioner, Solutions Architect – Associate, Machine Learning Engineer – Associate or Data Engineer – Associate. If you are newer to AI on AWS, the AWS AI Practitioner (AIF-C01) study guide covers the vocabulary this exam assumes, and the ML Engineer Associate study guide covers the SageMaker side.

The five domains (AWS’s published outline)

These weights come straight from the exam guide. The question counts are our own arithmetic on the 65 scored questions, so treat them as approximate.

Domains 1 and 2 together are 57% of the scored exam. If you are short on time, that is where the hours go. Below is what each domain covers, summarized from the task statements in the published guide. AWS notes the guide is not a complete list of exam content.

Domain 1: Foundation model integration, data management, and compliance (31%)

The biggest domain, and the one most people think of as “building a RAG app properly.” The guide’s six tasks:

• Design GenAI solutions: architectures that fit business and technical constraints, proofs of concept on Amazon Bedrock, and standard components using the AWS Well-Architected Framework and its Generative AI Lens.
• Select and configure foundation models: choosing models with benchmarks and capability analysis, switching models or providers without code changes (AWS Lambda, Amazon API Gateway, AWS AppConfig), resilience with AWS Step Functions circuit breakers and Amazon Bedrock Cross-Region Inference, and deploying fine-tuned models with Amazon SageMaker AI, LoRA and adapters, and the SageMaker Model Registry.
• Data validation and processing: data quality checks (AWS Glue Data Quality, SageMaker Data Wrangler), multimodal inputs (text, image, audio, tabular), and formatting requests the way each model expects.
• Vector stores: Amazon Bedrock Knowledge Bases, Amazon OpenSearch Service, Aurora and DynamoDB patterns, metadata for better precision, indexing at scale, and keeping the store current as source documents change.
• Retrieval: chunking strategies, choosing embeddings (Amazon Titan and other Bedrock embedding models), hybrid keyword-plus-vector search, Bedrock reranker models, query expansion and decomposition, and Model Context Protocol (MCP) clients for vector queries.
• Prompt engineering and governance: Amazon Bedrock Prompt Management, Amazon Bedrock Guardrails, Bedrock Prompt Flows for chained prompts, audit trails with AWS CloudTrail, and regression-testing prompts.

Domain 2: Implementation and integration (26%)

• Agentic AI and tools: agents with memory and state (Strands Agents, AWS Agent Squad, MCP), ReAct-style reasoning with Step Functions, safeguards such as stopping conditions, timeouts and IAM boundaries, human review steps, and MCP servers on Lambda or Amazon ECS.
• Model deployment: on-demand invocation versus Bedrock provisioned throughput versus SageMaker AI endpoints, the memory and GPU realities of LLM hosting, and cascading routine requests to smaller models.
• Enterprise integration: connecting to legacy systems, event-driven designs with Amazon EventBridge, identity federation and least-privilege access, data residency with AWS Outposts or AWS Wavelength, and CI/CD for GenAI components with AWS CodePipeline and AWS CodeBuild behind a central GenAI gateway.
• Foundation model APIs: synchronous calls, asynchronous processing with Amazon SQS, streaming responses, exponential backoff and rate limiting, AWS X-Ray tracing, and routing requests to the right model.
• Application patterns and developer tools: AWS Amplify front ends, OpenAPI-first design, Amazon Bedrock Data Automation for document workflows, Amazon Q Developer, and CloudWatch Logs Insights for troubleshooting.

Domain 3: AI safety, security, and governance (20%)

• Input and output safety: Bedrock guardrails on both prompts and responses, custom moderation workflows, grounding answers in a knowledge base to cut hallucinations, JSON Schema for structured output, and detecting prompt injection and jailbreak attempts.
• Data security and privacy: VPC endpoints, IAM, AWS Lake Formation, PII detection with Amazon Comprehend and Amazon Macie, masking and anonymization, and retention with S3 Lifecycle rules.
• Governance and compliance: programmatic model cards in SageMaker AI, data lineage and source tracking with AWS Glue, CloudTrail audit logs, and continuous monitoring for misuse, drift and policy violations.
• Responsible AI: explainable outputs and Bedrock agent tracing, fairness evaluations, A/B testing with Prompt Management and Prompt Flows, and LLM-as-a-judge evaluations.

Domain 4: Operational efficiency and optimization (12%)

• Cost: token tracking and context-window trimming, tiered model use by query complexity, provisioned throughput sizing, and caching (semantic caching and prompt caching).
• Performance: latency-optimized Bedrock models, response streaming, batch inference, and choosing temperature and top-k/top-p for the job.
• Monitoring: Amazon CloudWatch for token usage and response quality, Bedrock model invocation logs, tool-call and multi-agent tracing, vector store health, and golden datasets for spotting hallucinations.

If you build on AWS for real while you study, set a budget alert before you start making model calls. Our guide to budgets and alerts for usage-based AI spend walks through it.

Domain 5: Testing, validation, and troubleshooting (11%)

• Evaluation: measuring relevance, factual accuracy, consistency and fluency, Amazon Bedrock Model Evaluations, A/B and canary tests, LLM-as-a-judge, RAG and retrieval quality checks, agent task-completion measures, and validation gates before deployments.
• Troubleshooting: context window overflow and truncation, API integration errors, prompt problems, and retrieval issues such as poor embeddings or bad chunking.

The AWS services to know

AWS publishes a long in-scope services list, more than 100 services and features. You do not need equal depth in all of them. Based on how often they appear in the task statements, these deserve the most time:

• Amazon Bedrock and its features: Knowledge Bases, Guardrails, Prompt Management, Prompt Flows, model evaluations and provisioned throughput.
• Amazon Bedrock AgentCore, Strands Agents and MCP for agents. AWS refreshed the standard exam in March 2026 to reflect AgentCore, so do not rely on prep material older than that.
• Amazon SageMaker AI for hosting and managing customized models (Model Registry, Clarify, Model Monitor, JumpStart).
• Retrieval stores: Amazon OpenSearch Service, Amazon Aurora (pgvector), Amazon DynamoDB.
• Integration services: AWS Lambda, AWS Step Functions, Amazon API Gateway, Amazon EventBridge, Amazon SQS.
• Observability and security: Amazon CloudWatch, AWS X-Ray, AWS CloudTrail, IAM, AWS KMS, Amazon Macie, Amazon Comprehend.
• Developer tooling: Amazon Q Developer and Kiro.

A six-week study plan

AWS does not publish a recommended study time for AIP-C01. Our plan assumes you roughly match AWS’s target candidate and can give it about 10 hours a week, which works out to about 60 hours. If you have never built on Amazon Bedrock, add two weeks of hands-on time before you start. Here is where the hours go, split by domain weight:

1. Week 1: read the whole exam guide, then take AWS’s free 20-question Official Practice Question Set cold as a baseline. Start Domain 1 with design, model selection and data pipelines.
2. Week 2: finish Domain 1: vector stores, retrieval and prompt governance. Build one small RAG app on Bedrock Knowledge Bases, then change the chunking, embedding and reranking settings and watch what happens to the answers.
3. Week 3: Domain 2. Build a simple agent that calls one tool through MCP, then add a stopping condition and a timeout. Learn the deployment options side by side and the API patterns: streaming, asynchronous with SQS, backoff and routing.
4. Week 4: Domain 3. Put guardrails on input and output, practice PII detection and masking, and know which control (IAM, VPC endpoints, Lake Formation, Macie, CloudTrail) solves which described problem.
5. Week 5: Domains 4 and 5: caching, token budgets, monitoring, evaluation methods and troubleshooting. They are only 23% together, but the task statements are detailed.
6. Week 6: take the AWS Certification Official Practice Exam in one timed sitting with no notes, review every miss by domain, then re-read the task statements for your two weakest domains.

Short on hours? Trim Domains 4 and 5 first and protect Domain 1. And keep the out-of-scope list in mind: time spent on training math or feature engineering is time this exam will not reward. For a feel of how we pace a foundational AWS exam, compare our three-week AIF-C01 plan.

Exam rules that change how you study

• About 2.4 minutes per question. 180 minutes for 75 questions. Practice reading each scenario for the one constraint that decides the answer: cost, latency, compliance or least operational effort.
• Multiple response is all or nothing. The guide says you must select all the correct responses to get credit. Half right scores zero.
• Guess rather than skip. Unanswered questions are scored as incorrect and there is no penalty for guessing.
• One overall score. Scoring is compensatory, so a weak domain can be offset by strong ones. You only need 750 on the whole exam.
• Ten questions do not count, and you will not know which. An odd-looking question is not a reason to panic.
• Results take up to 5 business days, and a failed attempt means a 14-day wait and another full-price fee.

How AIP-C01 compares with the other AWS AI certifications

AIP-C01 is the most expensive and has the highest bar of the four. It also pulls the most weight at renewal time: AWS’s recertification page lists passing AIP-C01 as a way to recertify the AI Practitioner, the Machine Learning Engineer – Associate and the Data Engineer – Associate, and any Professional-level exam also recertifies Cloud Practitioner. If you are still deciding between them, see whether the ML Engineer Associate is worth it, what is changing in MLA-C02 and our ranking of AI certifications.

Official prep resources

• The exam guide (free). The task statements above are the closest thing to a syllabus.
• The Exam Prep Plan on AWS Skill Builder, which AWS links from the certification page.
• AWS Certification Official Practice Question Set: 20 exam-style questions, free on Skill Builder.
• AWS Certification Official Pretest and Official Practice Exam: AWS lists both in the prep plan. AWS says its Official Practice Exams need a Skill Builder subscription, which starts at 29 USD a month.
• Hands-on: AWS also points to AWS Builder Labs, AWS Cloud Quest, AWS Jam and AWS SimuLearn.

Skip exam dump sites. The AWS Certification Program Agreement, which you accept when you book, bars using leaked exam content, including material posted on third-party websites. Dumps are also often wrong, and they teach you answers instead of the judgment a professional-level exam tests. Our take on why dumps are a bad bet for AWS AI exams applies here too. When you are ready to plan the full cost, see what an AI certification actually costs or run the numbers in the AI certification cost calculator.

HOW TO // AI is not affiliated with or endorsed by Amazon Web Services. AWS Certified Generative AI Developer – Professional and AIP-C01 are certifications of Amazon.com, Inc. or its affiliates; we reference them descriptively. All content is original and based on AWS’s published exam guide. Exam details can change, so check the official exam page before you book.

Keep exploring: AIP-C01 exam cost · AWS AI certifications · generative AI certifications

Related guides

• AIP-C01 Exam Cost: AWS Generative AI Developer Professional Price, Retakes and Renewal
• AWS AI Certifications in 2026: Every Exam, the Path and the Costs
• AWS AI Practitioner (AIF-C01) Study Guide & Cheat Sheet
• MLA-C02 Exam Guide: What’s Changing From MLA-C01