AIP-C01 Exam Guide: Scope, Skills, Preparation Strategy, and Study Roadmap
AIP-C01 is the AWS Certified Generative AI Developer – Professional exam. It validates the ability to integrate foundation models into applications and business workflows, then operate those solutions with appropriate security, governance, testing, and cost controls. This guide is for experienced developers, cloud engineers, and AI practitioners deciding whether their background matches the professional-level target, which domains deserve the most study time, and whether they are ready to schedule the exam rather than rely on memorization or unofficial question collections.
What AIP-C01 actually validates
AIP-C01 validates production-oriented generative-AI development on AWS, not merely familiarity with AI terminology. The exam is intended for people performing a GenAI developer role and covers architecture, foundation-model integration, application implementation, responsible use, operational efficiency, and troubleshooting.
AWS describes the certification as evidence that a candidate can integrate foundation models into applications and business workflows. The stated task areas include vector stores, Retrieval Augmented Generation (RAG), knowledge bases, prompt engineering and management, agentic AI, security, governance, responsible AI, monitoring, optimization, and foundation-model evaluation.
That scope makes AIP-C01 different from an introductory AI survey. You should be able to reason about design choices in a production context: how data reaches a model, how retrieval improves an answer, how access is controlled, how outputs are evaluated, and how an application is monitored after deployment. The exam is not presented as a test of training a new model from first principles.
The intended candidate profile
AWS identifies a target candidate with 2 or more years of experience building production-grade applications on AWS or with open-source technologies, general AI/ML or data-engineering experience, and 1 year of hands-on experience implementing GenAI solutions. These are official target-profile indicators, not a stated prerequisite or guarantee of readiness.
AWS also recommends knowledge of compute, storage, and networking; security and identity management; deployment and infrastructure-as-code tools; monitoring and observability; and cost optimization. A candidate who knows only model prompting but cannot connect that work to IAM, data storage, deployment, and operations should treat those gaps as material preparation priorities.
What is outside the intended scope
The official guide lists model development and training, advanced ML techniques, and data engineering and feature engineering as out-of-scope job tasks for the target candidate. That does not mean data quality or model behavior can be ignored. It means preparation should emphasize integrating and operating GenAI solutions rather than becoming a specialist in training algorithms or advanced statistical modeling.
Check your readiness before you buy a study plan
A practical readiness check is whether you can explain an end-to-end GenAI solution and defend its trade-offs. Before scheduling, map your experience against model selection, retrieval, application integration, IAM, data protection, observability, cost, evaluation, and failure handling. If several of these are theoretical only, study the missing capability rather than starting with random practice questions.
Use the following self-assessment as a decision tool. For each item, write a short explanation in your own words and identify the AWS services or patterns that would support it:
• Select and configure an appropriate foundation model for a stated application need. • Explain when RAG, a knowledge base, or another retrieval design is appropriate. • Describe how documents become usable retrieval data, including validation and processing considerations. • Apply prompt-engineering and prompt-management techniques to a defined behavior problem. • Integrate a model or agent into an application or business workflow. • Protect data and credentials with suitable identity, encryption, network, and logging controls. • Monitor quality, latency, failures, usage, and cost after deployment. • Evaluate outputs for quality, safety, and responsibility rather than treating fluent text as correct. • Diagnose whether a failure originates in the prompt, retrieved context, model choice, application integration, or operational layer.
If your answers are lists of product names without explaining selection criteria, your weakness is probably architectural reasoning. If you can design the solution but cannot identify the security or operational controls, study the supporting AWS fundamentals before concentrating on advanced GenAI features. If you have little production application experience, allow additional time for implementation practice and scenario analysis.
Choose the right certification target
Do not confuse AIP-C01 with AIF-C01. AWS identifies AIP-C01 as AWS Certified Generative AI Developer – Professional, while AIF-C01 is AWS Certified AI Practitioner. The professional exam is aimed at integrating foundation models into production applications and workflows; the practitioner exam is designed for foundational AI concepts and AWS AI tools. A candidate seeking AIP-C01 should use the AIP-C01 exam guide and blueprint, not a practitioner study outline.
Understand the exam format and scoring
The exam contains 75 questions presented as multiple-choice or multiple-response items. AWS states that 65 questions affect the score and that 10 questions are unscored. Unscored questions are included for evaluation and do not affect the result, but you cannot identify them in advance, so every item should receive a considered response.
The result is reported as a scaled score of 100–1,000, and the minimum passing score is 750. AWS states that unanswered questions are scored as incorrect and that there is no penalty for guessing. Consequently, leaving an item blank is not a useful time-management tactic: mark the best answer after eliminating unsupported alternatives.
Multiple-choice items have one correct response and three distractors. Multiple-response items have two or more correct responses out of five or more options, and all correct responses must be selected to receive credit. Read the request for the number or type of required choices carefully. A response that is generally true may still be wrong if it does not satisfy the stated security, cost, latency, data, or business constraint.
The official AIP-C01 materials describe multiple-choice and multiple-response questions. Do not transfer the ordering and matching formats listed in the AIF-C01 guide to AIP-C01 preparation; those formats belong to the practitioner exam information supplied in the research, not the AIP-C01 evidence.
Use the format to improve decisions, not to memorize patterns
For each scenario, first identify the requested outcome and the constraints. Then separate requirements from attractive but irrelevant features. For example, a question may mention a foundation model, a document collection, an application workflow, and a security requirement. The correct choice will depend on the combination of retrieval design, integration method, access control, data handling, or operational objective—not on recognizing the most familiar service name.
Flag questions that depend on a subtle assumption and return to them later. On a multiple-response item, test every option independently against the scenario. Do not select an option simply because it is a valid AWS practice in a different architecture.
Allocate study time by the official blueprint
The blueprint gives the largest share to Applications of Foundation Models, so preparation should devote substantial practical attention to integration, retrieval, prompts, agents, and application behavior. The remaining domains are not optional: security, governance, responsible AI, testing, and operations often determine whether a technically plausible design is acceptable in production.
The AIP-C01 content domains and their weightings are:
• Content Domain 1: Foundation Model Integration, Data Management, and Compliance (31% of scored content). • Content Domain 2: Implementation and Integration (26% of scored content). • Content Domain 3: AI Safety, Security, and Governance (20% of scored content). • Content Domain 4: Operational Efficiency and Optimization for GenAI Applications (12% of scored content). • Content Domain 5: Testing, Validation, and Troubleshooting (11% of scored content).
The percentages describe scored-content weighting, not a promise that a particular sitting will contain a fixed sequence of topics. Use them to prioritize study and review, while still covering every domain. A common mistake is to focus almost exclusively on model APIs because they seem central to GenAI. That approach neglects the security, governance, testing, and operational judgments required by the professional-level blueprint.
Domain 1: Foundation Model Integration, Data Management, and Compliance
Domain 1: Foundation Model Integration, Data Management, and Compliance accounts for 31% of scored content and should anchor the study plan. Prepare to reason from a business requirement to an architecture: choose and configure a foundation model, plan data validation and processing, select a vector-store approach, design retrieval, and apply prompt-engineering and governance considerations.
Build a decision table rather than a product glossary. Record the application requirement, data source, retrieval need, model behavior, compliance concern, and likely AWS integration for each practice scenario. Include Amazon Bedrock, Amazon Bedrock Knowledge Bases, Amazon Bedrock Prompt Management, Amazon Bedrock Prompt Flows, Amazon Bedrock AgentCore, Amazon S3, and relevant application services where they fit the scenario.
Study the complete retrieval path. A useful mental sequence is source data, ingestion and processing, chunking or representation choices, indexing, retrieval, context assembly, model invocation, response handling, and evaluation. The exam may test the consequence of a design choice at any point in that chain. Knowing that a service exists is less valuable than understanding why a design would use it and what could fail.
Domain 2: Implementation and Integration
Domain 2: Implementation and Integration accounts for 26% of scored content and tests how GenAI capabilities become usable application features. Study how a model, retrieval component, agent, workflow, and ordinary application services interact, including authentication, orchestration, API boundaries, asynchronous work, error handling, and deployment concerns.
Connect GenAI study to general AWS engineering. The in-scope list includes AWS Lambda, Amazon ECS, Amazon EKS, AWS Step Functions, Amazon API Gateway, Amazon EventBridge, Amazon SQS, Amazon SNS, AWS AppSync, Amazon EC2, AWS CDK, AWS CloudFormation, AWS CodeBuild, AWS CodeDeploy, and AWS CodePipeline. Do not attempt to memorize every listing equally. Learn the role each category can play in an integrated solution and the constraints that make one choice more suitable than another.
Practice drawing a request path. Start with the client, identify the identity boundary, show the application component, model or agent call, retrieval or tool call, data stores, logging, and response path. Then mark where retries, timeouts, validation, authorization, and sensitive-data controls belong. This turns abstract service knowledge into architecture reasoning.
Domain 3: AI Safety, Security, and Governance
Domain 3: AI Safety, Security, and Governance accounts for 20% of scored content. Treat safety as a design property, not a final checklist. Preparation should cover access control, encryption, secrets, network boundaries, logging, data exposure, prompt or instruction risks, harmful outputs, policy enforcement, auditability, and responsible operation.
Review IAM, IAM Access Analyzer, AWS KMS, AWS Secrets Manager, Amazon Macie, AWS WAF, Amazon Cognito, AWS CloudTrail, Amazon CloudWatch, and related controls in context. For each control, ask what it protects, where it applies, what it does not solve, and which operational evidence it produces. This avoids the common mistake of choosing encryption as a complete answer to an authorization or model-behavior problem.
When reading a safety scenario, identify the asset and the failure mode before selecting a service. A model producing an unsafe answer, an application exposing a secret, an unauthorized user retrieving confidential documents, and an administrator needing an audit trail are different problems. They may require different preventive and detective controls.
Domain 4: Operational Efficiency and Optimization for GenAI Applications
Domain 4: Operational Efficiency and Optimization for GenAI Applications accounts for 12% of scored content. Prepare to balance quality, latency, reliability, throughput, and cost against the business requirement. A technically impressive model is not automatically the right production choice if it creates unacceptable response time or operating expense.
Study monitoring and cost as parts of the same operating picture. Relevant in-scope services include Amazon CloudWatch, Amazon CloudWatch Logs, AWS X-Ray, AWS Cost Explorer, AWS Cost Anomaly Detection, AWS Auto Scaling, and the AWS Well-Architected Tool. Pair each with a question: what signal is collected, which layer does it describe, how does it reveal a problem, and what action could follow?
Build comparison exercises around trade-offs. For a given workload, consider model selection, prompt size, retrieved-context size, caching or reuse where appropriate, concurrency, scaling, observability detail, and failure recovery. Avoid unsupported assumptions about a service's cost or performance. The exam expects selection reasoning from the scenario, not universal claims that one option is always cheapest or fastest.
Domain 5: Testing, Validation, and Troubleshooting
Domain 5: Testing, Validation, and Troubleshooting accounts for 11% of scored content. It is smaller by weighting but critical to production readiness. Study how to define expected behavior, evaluate response quality and responsibility, test retrieval and prompts, distinguish application failures from model failures, and use monitoring evidence to isolate regressions.
Create a fault-isolation worksheet. For a poor answer, ask whether the source data is incomplete, the retrieved context is irrelevant, the prompt is ambiguous, the model is unsuitable, the application passed the wrong input, a tool failed, or a safety control altered the response. For each cause, name the evidence you would inspect and the corrective action you would test.
Do not treat a fluent answer as a validated answer. A useful evaluation plan connects the test case to the business requirement and checks more than wording. Consider factual grounding, relevance, refusal behavior, sensitive-data handling, consistency, latency, and failure recovery where the scenario calls for them.
Build an AWS service map without memorizing a catalogue
Use the official in-scope service list to organize revision by architectural role. AWS says the list is non-exhaustive and subject to change, so it is a scope indicator rather than a substitute for the current exam guide. Study the services that support the blueprint and your weak areas, then verify changes against the official AIP-C01 materials before scheduling.
A useful service map has these layers:
• Model and GenAI capabilities: Amazon Bedrock, Amazon Bedrock AgentCore, Amazon Bedrock Knowledge Bases, Amazon Bedrock Prompt Management, Amazon Bedrock Prompt Flows, Amazon SageMaker AI, Amazon SageMaker JumpStart, Amazon SageMaker Clarify, Amazon SageMaker Model Monitor, Amazon SageMaker Model Registry, Amazon Comprehend, Amazon Kendra, Amazon Lex, Amazon Q Business, Amazon Q Developer, Amazon Titan, Amazon Textract, and Amazon Transcribe. • Data and retrieval: Amazon S3, Amazon OpenSearch Service, Amazon Aurora, Amazon DynamoDB, Amazon Neptune, Amazon RDS, Amazon ElastiCache, AWS Glue, Amazon Athena, Amazon EMR, Amazon Kinesis, and Amazon MSK. • Application and orchestration: AWS Lambda, Amazon EC2, Amazon ECS, Amazon EKS, AWS Fargate, AWS Step Functions, Amazon EventBridge, Amazon SQS, Amazon SNS, Amazon API Gateway, and AWS AppSync. • Delivery and operations: AWS CDK, AWS CLI, AWS CloudFormation, AWS CodeArtifact, AWS CodeBuild, AWS CodeDeploy, AWS CodePipeline, Amazon CloudWatch, Amazon CloudWatch Logs, AWS X-Ray, AWS Systems Manager, AWS Auto Scaling, and the AWS Well-Architected Tool. • Security and network controls: IAM, IAM Access Analyzer, IAM Identity Center, Amazon Cognito, AWS KMS, AWS Encryption SDK, AWS Secrets Manager, Amazon Macie, AWS WAF, Amazon VPC, AWS PrivateLink, Amazon CloudFront, Elastic Load Balancing, and Amazon Route 53.
For every service on your personal list, write one sentence describing its architectural role and one sentence naming a nearby alternative or boundary. For example, distinguish a model integration service from a compute service, a retrieval store from an object store, an identity control from an encryption control, and a monitoring service from a testing method. Those distinctions are more useful than isolated flashcards.
Keep service names tied to decisions
A good revision note answers four questions: what problem does the service address, what input or output does it handle, what constraint makes it appropriate, and what other component must connect to it? If a note only says “Service X is used for AI,” rewrite it. Scenario questions reward precise boundaries and consequences.
The official list includes services across analytics, application integration, compute, containers, customer engagement, databases, developer tools, machine learning, management and governance, migration and transfer, networking and content delivery, security, identity and compliance, and storage. That breadth is a reminder that AIP-C01 is an AWS application-development exam with GenAI at its center, not a narrow model-only test.
A practical study roadmap
A staged roadmap works better than reading every service page in sequence. First establish the blueprint and your gaps, then study architecture patterns, then implement or trace representative workflows, and finally test decision quality under time pressure. Keep an error log throughout; it should record the reason an answer was wrong, not just the correct option.
Use this sequence and adjust the amount of work to your existing experience:
1. Baseline and scope: Read the current official AIP-C01 exam guide, copy the five domains into a tracking sheet, and mark each topic as confident, familiar, or weak. Confirm that your materials are for AIP-C01 rather than AIF-C01. 2. Architecture foundation: Review AWS compute, storage, networking, IAM, deployment, observability, and cost concepts. Draw a simple GenAI application with a model call, data source, retrieval layer, application API, identity boundary, logs, and operational metrics. 3. Domain 1 focus: Work through foundation-model selection, data processing, vector stores, retrieval, knowledge bases, prompt engineering, prompt management, and compliance. For each topic, write a design choice and a failure mode. 4. Domain 2 focus: Trace model and agent integration into applications and workflows. Practice choosing between synchronous and workflow-oriented designs, identifying integration boundaries, and placing authentication, validation, retries, and monitoring. 5. Domain 3 focus: Review responsible AI, safety controls, identity, encryption, secrets, network isolation, data discovery, audit trails, and governance. Revisit each control through a concrete threat or policy requirement. 6. Domains 4 and 5 focus: Create operational scenarios involving latency, cost, quality regression, retrieval errors, failed tools, unsafe output, and incomplete monitoring. Decide what to measure and how to isolate the fault. 7. Mixed review: Alternate domains instead of studying only one service family at a time. Explain why each distractor is unsuitable. Re-read the official content outline for topics you cannot explain without notes. 8. Final readiness check: Take a timed, reputable practice assessment that does not claim to reproduce live questions. Review every uncertain response, verify changing service details in official documentation, and schedule only when your reasoning is consistent across domains.
Keep a single-page architecture vocabulary sheet, but do not turn it into a list of guessed exam answers. Include concepts such as foundation model, embedding, vector store, RAG, knowledge base, agent, prompt management, evaluation, guardrail or safety control, observability, governance, and least privilege. Define each concept in relation to a design decision.
If you have strong AWS experience but limited GenAI work
Start with model invocation, retrieval, prompt behavior, agents, evaluation, and GenAI-specific failure modes. Your cloud fundamentals may already cover IAM, networking, deployment, and monitoring, but do not assume that knowledge automatically explains model quality or retrieval behavior. Build small, controlled workflows and document what changes when the prompt, context, model, or tool changes.
If you have GenAI experience outside AWS
Translate familiar patterns into AWS service boundaries. Map your existing model provider, vector store, orchestration layer, identity system, secrets process, deployment pipeline, and monitoring stack to possible AWS equivalents. Then revisit the trade-offs: an equivalent name does not imply identical permissions, integration behavior, operational signals, or governance responsibilities.
If you mostly use AI tools rather than build them
Treat the target profile as a warning to strengthen implementation skills before scheduling. Study how an application invokes a model, retrieves authorized context, handles errors, records evidence, and controls cost. Reading conceptual summaries alone is unlikely to close the gap between using an AI feature and designing a production-ready GenAI solution.
Use hands-on work to test architectural understanding
Hands-on practice should produce explanations and evidence, not just a successful demo. Build or inspect a small workflow that stores source documents, retrieves relevant context, invokes a foundation model, applies access controls, records useful telemetry, and evaluates outputs. Then deliberately introduce an error and explain how you would locate it.
A productive lab sequence is:
• Define a business task and its unacceptable outcomes before selecting a model. • Store a controlled document set and identify its ownership, sensitivity, and update path. • Add retrieval and inspect whether the returned context actually supports the answer. • Change the prompt and observe whether the behavior improves for the stated evaluation criteria. • Add identity and least-privilege permissions for the application components. • Record request, response, latency, error, and usage signals without exposing sensitive content unnecessarily. • Test an incomplete source, an unauthorized request, an irrelevant question, and a model or tool failure. • Write a short review covering quality, safety, cost, performance, and the next improvement.
The point is not to reproduce a hidden exam task. It is to practice the reasoning the blueprint describes. If an experiment uses a service that changes or is unavailable in your environment, document the intended architecture from current AWS documentation and focus on the interface, control, and trade-off.
What to write after each lab
Record the requirement, architecture, assumptions, security boundary, evaluation method, observed failure, and alternative design. This creates revision material that is richer than screenshots. It also reveals whether your understanding is procedural—following steps—or transferable to a new scenario with different data, latency, compliance, or cost constraints.
Avoid these preparation mistakes
The most damaging mistakes are misidentifying the exam, studying product names without decisions, ignoring non-model domains, and trusting unofficial material as if it were authoritative. Correct them by returning to the blueprint, explaining trade-offs, and using official AWS sources for scope and current details.
Common pitfalls include:
• Preparing for AIF-C01 because its code is similar. Verify the title and code on every study resource. • Treating AIP-C01 as a model-training exam. The official out-of-scope tasks include model development and training and advanced ML techniques. • Memorizing a service list without understanding integration. Link each service to a requirement, boundary, and failure mode. • Assuming RAG automatically produces grounded or safe answers. Retrieval quality, authorization, prompt construction, source quality, and evaluation still matter. • Selecting encryption when the scenario asks for authorization, secrets management, auditability, or output safety. Identify the actual control objective. • Ignoring cost and operations because they have lower blueprint weightings. Domain 4: Operational Efficiency and Optimization for GenAI Applications is still tested, as is Domain 5: Testing, Validation, and Troubleshooting. • Treating practice-question scores as proof of readiness. Use them to locate reasoning gaps, not to predict an official result. • Leaving questions unanswered. AWS states that unanswered questions are scored as incorrect and that there is no penalty for guessing. • Relying on dumps, leaked questions, or memorization claims. Such material is not a substitute for understanding and may be inaccurate or unauthorized.
When reviewing a wrong answer, classify the error: knowledge gap, requirement misread, service-boundary confusion, security oversight, or premature selection. The category tells you what to do next. Re-reading a product definition will not fix a habit of overlooking the word “least,” “most cost-effective,” “authorized,” or “production.”
Scheduling and delivery details to verify
AWS lists AIP-C01 as a Professional-level certification exam with a 180 minutes exam duration, a listed cost of 300 USD, and delivery at a Pearson VUE testing center or through online proctoring. AWS lists English, Japanese, Korean, and Simplified Chinese as exam languages. Confirm the current registration, delivery, pricing, and policy details on the official certification page before booking because operational information can change.
Schedule after you have reviewed all five domains and can explain a complete architecture without depending on a memorized sequence. Before booking, check the current official page for appointment availability, candidate requirements, identification rules, rescheduling terms, and any delivery-specific technical checks. The supplied sources support the delivery options and listed language information, but they do not establish your local availability or personal eligibility.
AWS certifications are valid for three years from the date earned and require recertification to remain current and active. Treat that as a maintenance consideration when deciding whether the professional certification fits your role and near-term development plan; consult AWS recertification policy for the applicable route and current rules.
A sensible final-week routine
In the final week, stop expanding the service catalogue. Revisit your error log, blueprint, architecture diagrams, security decisions, evaluation methods, and operational trade-offs. Complete mixed scenario practice, practise selecting all correct responses on multiple-response items, and reserve time to verify official updates rather than filling gaps with unverified answer banks.
What to do next
Begin with the official AIP-C01 exam guide and blueprint, then make a gap-led plan. Your immediate objective is not to collect more questions; it is to prove that you can connect foundation models to applications while preserving data control, responsible behavior, measurable quality, and operational value.
Take these next actions:
1. Confirm that your target is AWS Certified Generative AI Developer – Professional (AIP-C01), not AWS Certified AI Practitioner (AIF-C01). 2. Read the current AIP-C01 exam guide and record the five domain labels and weightings. 3. Rate yourself against the target experience and recommended AWS knowledge without treating either as a formal prerequisite. 4. Choose one representative GenAI workflow and draw its data, model, retrieval, identity, deployment, monitoring, and evaluation paths. 5. Start an error log and use it for every practice session or hands-on experiment. 6. Study Domain 1: Foundation Model Integration, Data Management, and Compliance first if you need a priority starting point, then cover integration, safety and governance, operations, and testing. 7. Recheck official AWS pages immediately before scheduling for current scope, delivery, pricing, languages, and policy details.
AIP-C01 preparation is strongest when every product choice is tied to a requirement and every output is tested against quality, safety, security, cost, or operational evidence. That approach prepares you for unfamiliar scenarios without pretending that unofficial question collections can reproduce the live exam.
subsections
Conclusion
Use AIP-C01 as a readiness test for production GenAI engineering, not as a catalogue-recitation exercise. The official blueprint rewards breadth across foundation-model integration, application implementation, safety, operations, and troubleshooting, with the greatest emphasis on integration and data decisions. Confirm the current AWS documentation, close the gaps revealed by your architecture work and error log, and schedule only when you can justify design choices under realistic constraints.
