AWS Certified Generative AI Developer – Professional (AIP-C01)

Learning plan

About 100 hours of focused study, split into 5 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 10 weeks

What the exam covers

Source: official exam guide. Weights can change, so check before you start.

Before you start

• An AWS account with Amazon Bedrock model access enabled in at least one Region
• Solid Python or TypeScript plus the AWS SDK, Lambda, API Gateway, Step Functions and IaC (CDK, CloudFormation or Terraform)
• AWS recommends 2+ years building production apps and 1 year of hands-on GenAI. If you lack the GenAI year, build the labs below for real rather than reading about them

1. Weeks 1–3 · 30 h

   FM selection, vector stores and retrieval (largest domain)

   • Choose FMs on benchmarks, context window, modality and price; design for model switching with AppConfig and Lambda, and for outages with Bedrock Cross-Region Inference
   • Design a vector store (Bedrock Knowledge Bases, OpenSearch Service with the Neural plugin, Aurora pgvector) with metadata filtering and an incremental sync strategy
   • Compare fixed-size, hierarchical and semantic chunking and Titan embedding dimensions; add hybrid search, query decomposition and Bedrock reranker models
   • Govern prompts with Bedrock Prompt Management (parameterised templates, versions) and chain them with Bedrock Prompt Flows
   • Know when to fine-tune with LoRA and deploy through SageMaker Model Registry versus staying with RAG

   Hands-on: Build two Bedrock Knowledge Bases over the same document set, one with fixed-size and one with hierarchical chunking. Run 20 test questions through RetrieveAndGenerate with and without a reranker and record which setup answers more correctly.

   Use: AWS Skill Builder – AIP-C01 exam prep · Amazon Bedrock User Guide · Amazon Bedrock Workshop (GitHub)

2. Weeks 4–6 · 26 h

   Agents, tools and enterprise integration

   • Build an agent with Strands Agents and host it on Bedrock AgentCore (Runtime, Memory, Gateway, Identity); know when AWS Agent Squad fits for multi-agent routing
   • Expose tools through MCP, with Lambda for lightweight stateless servers and ECS for heavier ones, and add timeouts, stop conditions and circuit breakers with Step Functions
   • Stream responses with the Bedrock ConverseStream API over WebSockets or SSE, and add retries with exponential backoff and API Gateway throttling
   • Choose on-demand, provisioned throughput or SageMaker AI endpoints for FM hosting, and route routine queries to smaller models (model cascading)
   • Wire GenAI into existing systems with EventBridge, SQS and CodePipeline-based CI/CD for prompts and agents

   Hands-on: Write a Strands agent with two tools (one Lambda-backed MCP tool that queries DynamoDB, one Knowledge Base retrieval tool), deploy it to AgentCore Runtime with AgentCore Memory, and call it from a Lambda behind API Gateway with streaming enabled.

   Use: Amazon Bedrock AgentCore Developer Guide · Strands Agents SDK · AgentCore samples (GitHub) · AWS Agent Squad (GitHub)

3. Weeks 6–8 · 20 h

   Safety, security and governance

   • Configure Bedrock Guardrails (content filters, denied topics, PII redaction, contextual grounding checks) and apply them to both input and output
   • Defend against prompt injection and data leakage with input validation, least-privilege IAM per tool, VPC endpoints (PrivateLink) and KMS
   • Log and audit model calls with Bedrock invocation logging, CloudTrail and CloudWatch Logs; track lineage and model cards for compliance
   • Apply the Well-Architected Generative AI Lens and the Responsible AI Lens to a design review

   Use: Amazon Bedrock Guardrails · Well-Architected Generative AI Lens · AWS Securing Agent Identities Demonstrated (Skill Builder)

4. Weeks 8–9 · 14 h

   Cost, performance, evaluation and troubleshooting

   • Cut token spend with prompt caching, batch inference, smaller or distilled models and intelligent prompt routing, and estimate cost per request
   • Evaluate RAG and model quality with Bedrock Evaluations (LLM-as-a-judge, RAG evaluation) and build prompt regression tests into CI
   • Trace FM calls with X-Ray and CloudWatch Logs Insights, and diagnose throttling, context-window overflow, hallucination and retrieval misses

   Hands-on: Add a Bedrock Guardrail and invocation logging to your agent, then run a Bedrock RAG evaluation job against your Knowledge Base and use CloudWatch Logs Insights to find the five most expensive prompts by token count.

   Use: Amazon Bedrock model evaluation

5. Week 10 · 10 h

   Practice exams and gap review

   • Do the official practice question set and the Skill Builder exam-prep plan, then a full 75-question timed practice exam in 180 minutes
   • Map every miss to a task statement (for example 1.5 retrieval or 2.1 agents) and rebuild that piece in your lab account
   • Re-read the in-scope and out-of-scope service lists in the exam guide so you can rule out distractors quickly

   Use: AWS Skill Builder – AIP-C01 exam prep · Tutorials Dojo AIP-C01 practice exams

You're ready to book when…

• You can sketch a production RAG plus agent architecture on AWS, naming the service for chunking, embeddings, vector store, reranking, orchestration, guardrails and observability
• You can explain when to choose Strands on AgentCore vs Bedrock Agents vs Step Functions orchestration
• You can list three ways to cut Bedrock inference cost and when each one hurts quality
• You score 75% or higher on two different full-length practice exams

Exam tips

• The passing score is 750 of 1,000, higher than the 720 needed for Associate exams, and 10 of the 75 questions are unscored.
• Model training and feature engineering are explicitly out of scope, so when an option involves training a model from scratch it is usually a distractor.
• If you fail, AWS makes you wait 14 days before a retake and you pay the full fee again.