AI-200 Exam Guide 2026: Developing AI Cloud Solutions on Azure
AI-200 is the Azure developer exam rebuilt for AI workloads. Containers and messaging survived from AZ-204; what is new is vector search across Cosmos DB, PostgreSQL and Redis.
On this page
AI-200 — Developing AI Cloud Solutions on Azure — is the Azure developer associate exam rebuilt around AI workloads, succeeding AZ-204. Microsoft's own study guide is candid about the lineage: its “get trained” and Exam Readiness Zone links still point at the az-204 pages.
Built against the official study guide as published 5 May 2026.
What AI-200 certifies
You contribute to all phases of implementing AI solutions on Azure with an emphasis on back-end services and components, and support the whole development lifecycle — requirements, design, development, deployment, security and monitoring.
Microsoft expects proficiency in Azure and third-party SDKs, Azure data management services, monitoring and troubleshooting, messaging and eventing, vector databases, Python, and containerised applications on Azure.
AI-200 is not about prompting models. It is about building the platform underneath an AI application — where it runs, where its embeddings live, how services talk to each other, and how you secure and observe the result.
AI-200 versus AI-103
Microsoft shipped two Azure AI associate exams close together and they are easy to confuse. They are for different jobs.
| AI-103 | AI-200 | |
|---|---|---|
| Role | Azure AI Engineer — application layer | AI Cloud Developer — backend platform |
| Centre of gravity | Microsoft Foundry, agents, RAG, prompt engineering | Containers, data services, messaging, observability |
| Predecessor | AI-102 | AZ-204 |
| Typical question | “How do I ground this agent?” | “Where do these embeddings live and how do I scale the query?” |
| Take it if | You build AI applications and agents | You build and run the Azure services they depend on |
If you are choosing, ask which half of the stack you actually own. Our AI-103 guide covers the application side.
The four domains and their weights
Containerised solutions (20–25%)
Container application hosting — building, storing, versioning and managing images in Azure Container Registry, ACR Tasks, and deploying containers to App Service with environment variables and secrets. Then orchestration: Azure Container Apps including environment configuration and revision management, event-driven scaling with KEDA, deploying to AKS with manifest files, and troubleshooting via logs, events and end-to-end connectivity.
Data management services (25–30%)
The largest domain, and the most AI-specific. Covered in the next section.
Connect to and consume Azure services (20–25%)
Service Bus (dead-letter handling, messages, topics, subscriptions) and Event Grid (filters, custom events, retries) for event- and message-based solutions, plus Azure Functions — serverless APIs with triggers and bindings, and deploying function apps.
Secure, monitor, troubleshoot (20–25%)
Key Vault for secrets including rotation and retrieval, App Configuration for app settings, distributed tracing with OpenTelemetry SDKs, and writing KQL queries to analyse logs and metrics. OpenTelemetry and KQL are both worth real practice — they are easy to recognise and hard to answer precisely.
The vector-database content that is new
Three data services, each with its own vector story. This is the clearest departure from AZ-204.
Given three services that can all store vectors, expect to be asked which one fits a scenario. Rough heuristic: Cosmos DB for operational documents at scale with a change feed, PostgreSQL when you want relational modelling and metadata-filtered RAG via pgvector, Redis when latency and caching dominate.
What carried over from AZ-204
More than you might expect. Container Registry, App Service, Container Apps, AKS, Functions with triggers and bindings, Service Bus, Event Grid, Key Vault, App Configuration and Cosmos DB fundamentals were all AZ-204 territory. If you were partway through AZ-204 preparation, that work is not wasted.
What is genuinely new: the vector and embedding layer across all three data services, Azure Managed Redis specifically, KEDA-based event-driven scaling, and OpenTelemetry tracing.
Question patterns to expect
AI-200 is a developer exam, so most questions are “pick the right service or setting for this requirement”. A few recurring shapes:
| Pattern | What it sounds like | How to attack it |
|---|---|---|
| Which data service? | “Store embeddings for a RAG app that also filters on tenant and date” | Metadata-filtered RAG → PostgreSQL with pgvector. Pure low-latency lookup → Redis |
| Which compute? | “Scale to zero between bursts of queue traffic” | Scale-to-zero and event-driven → Container Apps with KEDA or Functions, not AKS |
| Messaging or eventing? | “Process the order once, with retries and a poison path” | Commands with dead-lettering → Service Bus. Broadcasting a state change → Event Grid |
| Cost or performance tuning | “Queries are slow and RU consumption keeps climbing” | Cosmos DB → indexing policy and consistency level. PostgreSQL → index strategy and pgvector overhead |
| Where is the failure? | “Requests fail intermittently across three services” | OpenTelemetry distributed tracing, then KQL over the logs |
The service-selection questions are where the marks concentrate. For every service in the objectives, be able to finish the sentence “you would choose this over the alternatives when…”. Knowing what a service does is not enough when three of them could technically do it.
A five-week study plan
The study-time calculator will size this against your real availability.
Frequently Asked Questions
What does AI-200 replace?
AI-200 is Azure's developer associate exam for AI cloud solutions, succeeding AZ-204. Microsoft's own study guide links the AI-200 training and video resources to the az-204 pages, which reflects the lineage.
What is the difference between AI-200 and AI-103?
AI-103 is the application-focused Azure AI Engineer credential built around Microsoft Foundry, agents and RAG. AI-200 is the backend-platform credential: containers, AI-ready data services, event-driven integration, and securing and monitoring solutions on Azure.
Do I need Python for AI-200?
Yes. The audience profile lists Python programming alongside Azure and third-party SDKs, data management services, monitoring and troubleshooting, messaging and eventing, vector databases, and containerised applications on Azure.
How much of AI-200 is about vector databases?
Vector work runs through the largest domain, Develop AI solutions by using Azure data management services, which is 25 to 30 percent. It covers vector similarity search in Cosmos DB, pgvector in Azure Database for PostgreSQL, and vector indexing in Azure Managed Redis.
Practise AI-200 Before You Book
500–1,000+ AI-200 practice questions with worked explanations across all four domains.
AI-200 Practice Test
Practice Before You Book
500–1,000+ practice questions per exam with detailed explanations, across Azure, AWS, GCP, security, and AI certifications.