Every scenario question on this exam starts with a business requirement and ends with a service choice. Module 1 builds the vocabulary and the working environment you will use for the next seven weekends.

• Python for Azure developers: async patterns, typing, project structure
• Azure SDK conventions, clients, credentials and retry behaviour
• Managed identity and Entra ID over connection strings and keys
• Resource groups, App Configuration and environment promotion
• Working in VS Code and the Azure CLI the way the exam assumes
• Hands on: scaffold and deploy your first Azure AI service call

This is the domain most candidates underestimate. It is pure implementation and it rewards hands on repetition.

• Azure Container Registry, image management and ACR Tasks for cloud builds
• Deploying containers to App Service with environment variables and secrets
• Azure Container Apps: environments, ingress, revisions and traffic splitting
• Event driven scaling with KEDA in Container Apps
• Deploying to Azure Kubernetes Service using manifest files
• Monitoring and troubleshooting AKS and Container Apps through logs and events
• Hands on: ship the same AI service to Container Apps and to AKS, then compare

The exam does not ask which database is best. It gives you a constraint and asks which one satisfies it, which is why we teach the trade offs rather than the feature lists.

• Cosmos DB for NoSQL: SDK access, queries, partitioning and the change feed
• Azure Database for PostgreSQL with pgvector for relational plus vector workloads
• Azure Managed Redis for caching, session state and low latency lookups
• Choosing between them: change feed ingestion, relational joins, latency and cost
• Consistency, throughput and the failure modes that show up in production
• Hands on: the same dataset modelled three ways, measured side by side

Vector work is the part of AI-200 that did not exist in AZ-204, and it is the part hiring managers care about most.

• How embeddings work and how to choose an embedding model
• Chunking strategies that decide whether retrieval succeeds or fails
• Vector similarity search across Cosmos DB, pgvector and Redis
• Azure AI Search: indexes, vectorization and hybrid retrieval
• Grounding a model on your own content with citations and freshness handling
• Measuring retrieval quality instead of guessing at it
• Hands on: a grounded assistant over a real document corpus

AI-200 leans on back end services rather than model theory. This module covers the AI surface you are expected to integrate and operate.

• Azure AI Foundry: model catalog, deployments and SDK access
• Language and speech: extraction, summarization, translation and transcription
• Vision and Document Intelligence for images, forms and invoices
• Function calling and structured output for reliable application behaviour
• Content safety filters and responsible AI controls in the request path
• Hands on: a document processing service with human review for low confidence results

AI calls are slow, rate limited and occasionally fail. The exam tests whether you know how to build around that reality.

• Azure Service Bus queues and topics for reliable work distribution
• Event Grid and Event Hubs for event driven and high throughput streams
• Azure Functions and Durable Functions for orchestration and long running work
• Retries, dead lettering, idempotency and exactly once thinking
• Backpressure and throttling when the model provider is the bottleneck
• Hands on: an event driven pipeline that survives failure and replays cleanly

The final domain is where scenario questions get decided, and it is also what separates a demo from a system somebody trusts in production.

• Key Vault, managed identity, private endpoints and secret rotation
• Application Insights, distributed tracing and OpenTelemetry instrumentation
• Quotas, throttling, token cost tracking and capacity planning
• Diagnosing the common failures: cold starts, timeouts, poisoned messages
• Full AI-200 skills outline reviewed in blueprint proportion
• A full length domain scored mock exam plus a score report review

The Anthropic side of the program. You rebuild part of your Azure work on Claude, which is the fastest way to understand where each platform actually wins.

• Claude platform mastery: models, Projects, artifacts and tool use
• Building with the Claude API and with Claude Code
• Prompt caching, batching and cost control at production volume
• Anthropic certification preparation with practice questions and a mock
• Capstone: one product shipped on Azure and rebuilt on Claude, with a written comparison
• Portfolio write up and a mock interview defending your architecture choices