AWS Certified Generative AI Developer – Professional
Code
AIP-C01
Length
180 minutes
Pass mark
750 out of 1,000
Price
$300
Valid for
3 years, the AWS standard
Study time, our estimate
90 to 130 hours
AIP-C01Professional
The heaviest GenAI exam in this guide, and one of the most respected. At three hours and 75 questions, stamina matters. It covers production work on Bedrock: RAG, agents, safety, cost and troubleshooting. Note what is out of scope: model training, advanced ML and feature engineering.
Take it if
You run GenAI in production on AWS, with a year or more of hands-on work.
Skip it if
You are new to AWS. It assumes solid knowledge of compute, networking, IAM and infrastructure as code.
Before you start
Check yourself against this list first. If something is missing, close that gap before you pay for an exam.
Two years of production appsAWS recommends this on AWS or open-source stacks.
One year of hands-on GenAIShip one RAG or agent project first.
Strong AWS foundationsCompute, networking, IAM, IaC and monitoring.
Model training is not requiredAWS lists training, advanced ML and feature engineering as out of scope.
75 questions, 65 scored, multiple choice and multiple response
Length
180 minutes
Pass mark
750 out of 1,000
Price
$300
Valid for
3 years, the AWS standard
Delivery
Pearson VUE, test center or online
Languages
English, Japanese, Korean, Simplified Chinese
Experience
2+ years of production apps, 1 year of hands-on GenAI
If you fail
Wait 14 days
What the exam covers
Foundation Model Integration, Data Management, and Compliance31%
Implementation and Integration26%
AI Safety, Security, and Governance20%
Operational Efficiency and Optimization for GenAI Applications12%
Testing, Validation, and Troubleshooting11%
What to focus on
Domain 1 is nearly a third: choosing models, data pipelines, vector stores, retrieval, and prompt governance.
Agents and tool integration, model deployment strategies, and enterprise integration.
Input and output safety, data privacy, and governance. A fifth of the exam.
Cost, performance and monitoring for GenAI specifically.
Evaluation systems, and troubleshooting a GenAI app.
Everything on one mapEvery topic and key idea in this certification.
mindmap
root((AIP-C01))
Models, data and retrieval
Provider switching
Vector stores
Retrieval
Prompt governance
Agents and integration
Agent frameworks
Safeguards
Stateless MCP
Resilience
Safety and governance
Defense in depth
Privacy
Traceability
Cost and performance
Token efficiency
Caching
Throughput
Testing and troubleshooting
Evaluation
Deployment checks
Troubleshooting
A guarded Bedrock requestThis path touches four of the five domains.
sequenceDiagram
participant U as User
participant A as API layer
participant G as Guardrails
participant K as Knowledge Base
participant M as Foundation model
U->>A: Question
A->>G: Check the input
G-->>A: Allowed
A->>K: Retrieve
K-->>A: Chunks
A->>M: Prompt with context
M-->>A: Answer
A->>G: Check the output
G-->>U: Filtered answer
Lesson 1 of 5, 31% of the exam
Models, data and retrieval
Nearly a third of the exam, across six tasks.
Provider switching
Design so models can change without code changes.
Vector stores
Choose, add metadata, and keep the store current.
Retrieval
Chunking, embeddings, and hybrid search with reranking.
Prompt governance
Use Prompt Management to control versions and approvals.
Inside a RAG pipelineTwo paths: one fills the store, one reads it.
flowchart LR
A["Documents"] --> B["Chunk"] --> C["Embed"] --> D[("Vector store")]
E["Question"] --> F["Embed"] --> D
D --> G["Rerank"]
G --> H["Foundation model"]
H --> I["Grounded answer"]
Common trap
Changing the model when the chunks are the problem.
Try it yourself
Build a Knowledge Base and tune chunking against a test set.
Covers agents, deployment, and the enterprise around them.
Agent frameworks
The guide names Strands Agents and AWS Agent Squad.
Safeguards
Step Functions can enforce stopping conditions and human review.
Stateless MCP
The guide mentions Lambda for lightweight MCP servers.
Resilience
Exponential backoff, rate limiting and fallbacks.
Agent or Step Functions?Audited steps want determinism.
flowchart TD
A{"Are the steps fixed and audited?"} -->|Yes| B["Step Functions"]
A -->|No| C{"Does it need tools and planning?"}
C -->|Yes| D["Agent"]
C -->|No| E["Single model call"]
D --> F["Add stopping conditions and human review"]
Common trap
Using an agent for an audited, fixed process.
Try it yourself
Build the same flow as an agent and as Step Functions.
Check both input and output, and ground responses to reduce hallucination.
Privacy
Comprehend and Macie can detect sensitive data, and VPC endpoints isolate traffic.
Traceability
Track data sources and keep model cards.
Defense in depthA bad request has to get through every layer.
flowchart LR
A["Input"] --> B["Guardrail check"]
B --> C["Grounding"]
C --> D["Model"]
D --> E["Output guardrail"]
E --> F["Logged response"]
B -.->|"Blocked"| G["Safe refusal"]
E -.->|"Blocked"| G
Common trap
Relying on system prompt instructions for policy.
Try it yourself
Add a guardrail and test it with injection attempts.
GenAI cost behaves differently from normal compute.
Token efficiency
Track tokens and tier models by complexity.
Caching
Prompt caching and semantic caching avoid repeat work.
Throughput
Cross-Region inference helps with bursts.
Avoiding the model callThe cheapest token is the one you never send.
flowchart TD
A["Request"] --> B{"Seen a similar question?"}
B -->|Yes| C["Semantic cache"]
B -->|No| D{"Repeated prefix?"}
D -->|Yes| E["Prompt cache"]
D -->|No| F{"Simple request?"}
F -->|Yes| G["Cheaper model"]
F -->|No| H["Strong model"]
Check context overflow, API errors, prompts and retrieval.
Finding the failing layerCheck what the model received before you change it.
flowchart TD
A["Wrong answer"] --> B{"Was the context right?"}
B -->|No| C["Fix retrieval"]
B -->|Yes| D{"Was it cut off?"}
D -->|Yes| E["Fix context window overflow"]
D -->|No| F["Fix the prompt or model choice"]
Common trap
Rewriting prompts before looking at what the model received.
Exam facts checked on October 31, 2026 against AIP-C01 official exam guide. Lessons are our own explanations, so use the official docs as the final word, and confirm price and dates before booking.
