AI-200: Azure AI Cloud Developer Associate
A condensed guide to Microsoft Exam AI-200, Developing AI Cloud Solutions on Azure, written from Microsoft Learn and organised around the published skills list.
The guide is free. An account keeps it in your dashboard, where each module opens in order.
What is inside
Introduction to AI-200
What Exam AI-200 is, the back-end developer Microsoft wrote it for, the four skill areas and where the marks sit, how a pass is worked out, what changed when it replaced AZ-204, and how to prepare while there is no practice assessment and no learning path we could find.
about 15 min
Container images and Azure Container Registry
What a container registry actually stores, what each Azure Container Registry pricing plan buys, how to tag and version images so a deployment is reproducible, how to stop a registry filling up with orphaned manifests, when an ACR Task beats building locally or in your own pipeline, and how a container host pulls an image without anyone storing a password.
about 21 min
App Service and Container Apps
Running the image you built somewhere it can serve traffic. Deploying a container to Azure App Service and getting environment variables and secrets into it without baking them into the image, deploying to Azure Container Apps with its environment, secrets and revisions, splitting traffic between revisions, and using KEDA scale rules so an app grows with its queue and costs nothing when the queue is empty.
about 22 min
AKS, and troubleshooting containers
What goes in a Kubernetes manifest file and the object kinds AI-200 expects, how kubectl applies and manages them on Azure Kubernetes Service, when AKS is warranted over Container Apps, and a disciplined order for diagnosing a failing workload through status, events, logs and hop by hop connectivity on both services.
about 18 min
Azure Cosmos DB for NoSQL
Connecting with the Python SDK and running queries, what makes a request cost more Request Units, indexing policies and consistency levels, storing embeddings and running vector similarity search, and the change feed with its processor. Partition key choice underlies all of it.
about 25 min
Azure Database for PostgreSQL
Connecting and querying from Python, modelling tables and choosing data types for embeddings, the three vector index types and what each costs, the knobs that trade recall for latency, sizing compute, memory and storage for vector work, similarity search and RAG with metadata filters, and connection pooling with the built in PgBouncer.
about 25 min
Azure Managed Redis
The two skills Microsoft measures on Redis in AI-200. Which Redis service is current and how it relates to Azure Cache for Redis, the decisions you can only make when the instance is created, cache-aside and the ways it fails, expiry and eviction against explicit invalidation, and vector indexing with RediSearch for similarity search and semantic caching.
about 17 min
Service Bus and Event Grid
The two messaging skills Microsoft measures in AI-200. The difference between a message and an event and why the exam keeps asking it, queues and peek lock, topics with subscription filters, every route into the dead-letter queue and how to get a message back out, then Event Grid custom events, the three ways to filter a subscription, and exactly what the retry and dead-letter settings do.
about 22 min
Azure Functions
The two Azure Functions skills Microsoft measures in AI-200. Why exactly one trigger starts a function, how input and output bindings remove the client code you would otherwise write, which trigger a described scenario implies, the hosting plans and what a cold start costs when the work is a call to a model, then how settings and secrets reach a running function app and how each plan is deployed to.
about 26 min
Key Vault and App Configuration
The two skills Microsoft measures on secrets and settings. What belongs in Azure Key Vault against what belongs in Azure App Configuration, retrieval through a managed identity rather than a connection string, how secret rotation is triggered and the two shapes it takes, Key Vault references as the join between the services, and the three ways a running application picks up a rotated secret without a redeploy.
about 24 min
OpenTelemetry and KQL
"The two monitoring skills Microsoft measures in AI-200. What a span is and why a trace answers questions a log cannot, the one call that turns the Azure Monitor OpenTelemetry Distro on in Python, custom spans and the table they land in, trace context carried across a service boundary, cloud role names, what sampling keeps and drops, then Kusto Query Language from the bar upwards: the mental model, the eight words that do most of the work, real queries read back in plain English, and the three query habits the exam rewards."
about 22 min
Final review and exam day
The revision page for the last week. All twenty-seven skill bullets in one line each, weighted the way Microsoft weights them, the scenario cues behind the decisions that are easiest to get wrong, the services you will meet under two names, the dates and preview flags that change an answer, what Microsoft publishes about the exam itself, and a plan for the final seven evenings.
about 25 min
How it was written
These study notes summarise Microsoft Learn material for Exam AI-200. Microsoft has not published a learning path for this exam, so the guide is organised around the published skills list instead, one module per group of skills. Every module is written from the current documentation rather than copied from it, and each was checked against Microsoft Learn a second time before publishing, by someone other than the writer. Each module says when it was last checked and links the Learn pages it drew on.
For the official list of skills measured, see the Microsoft Learn study guide for Exam AI-200. This is a developer exam, and a guide can only take you so far on one: several modules show real code, but the parts you have to write yourself are best learned by doing them.