AI-200 study scenarios: three example profiles
The right way to study for AI-200 depends on which half of its title you already know: the cloud developer half or the AI half. The exam is 120 minutes, scaled to 700 out of 1000, with four domains that are close in size, and the largest one, AI solutions on Azure data management services (25 to 30%), is the one with the least overlap with the exam it replaced, AZ-204.
Example profiles, not real people. The profiles below are illustrative planning examples built from the official exam outline. They are not reviews or reports of anyone’s result.
The study plan allows four weeks at 8 to 10 hours, about 32 to 40 hours, one week per domain. The profiles below show where to change that.
The candidate who studied AZ-204
Starting point. Knows Functions, Service Bus, Event Grid, Key Vault and App Service well. Containers on Container Apps and AKS may be new, and the data domain is a different subject from the old storage domain.
Where the time goes. Data management (25 to 30%) takes the most, because Cosmos DB vector policies, pgvector index choices and Redis vector indexing each get their own objectives. Containerized solutions (20 to 25%) takes the second share, mainly KEDA scaling and AKS manifests.
The trap. Assuming the middle two domains are the same exam as before and revising them from old notes. Much of the material online covers topics that were dropped, and it misses vector search entirely.
Order. Data management, containers, then a quick pass over connecting services and the security and monitoring domain, concentrating on OpenTelemetry and KQL.
Planning assumption. A reasonable planning assumption is 28 hours: about 10 on data, 7 on containers, 4 on connecting services, 4 on secure, monitor and troubleshoot, and 3 for practice.
The Python ML engineer who has only used notebooks and managed endpoints
Starting point. Knows embeddings, similarity and retrieval conceptually, and writes decent Python. Has rarely containerised anything or wired queues and events.
Where the time goes. Everything outside data: containers, messaging, Functions, Key Vault, App Configuration, KQL. These make up 60 to 75% of the paper between three domains.
The trap. Assuming this is a model exam. The outline has no model selection, prompting or agent frameworks, so skill with models earns almost nothing and the weighting punishes the gap in infrastructure.
Order. Containers first so there is somewhere to run code, then connecting services, then data (familiar concepts, new services), then security and monitoring.
Planning assumption. A reasonable planning assumption is six weeks at 8 to 10 hours, or about 50 hours, with roughly 12 on containers, 10 on connecting services, 10 on security and monitoring, 12 on data, and 6 on practice.
The database developer strong in PostgreSQL and Redis
Starting point. Comfortable with indexing, query plans, connection pooling and caching. Has not worked with Cosmos DB’s request units and consistency levels, or used Azure’s container and eventing services.
Where the time goes. The Cosmos DB for NoSQL section (RU consumption, indexing policy, the change feed processor), then all of containers and connecting services. The PostgreSQL and Redis objectives are the shortest stretch of study.
The trap. Going deep on the services already known. The domain weight is generous, but 20 to 25% of the exam sits in each of the other three, and any one of them is large enough to fail on.
Order. Cosmos DB first, containers, messaging and Functions, then secrets and telemetry, then a light pass on PostgreSQL and Redis.
Planning assumption. A reasonable planning assumption is 36 hours, with 8 on data (mostly Cosmos DB) and roughly 7 on each of the other three domains, and the rest on practice.
The three AI-200 profiles compared
| Profile | Front-loaded domains | Main risk |
|---|---|---|
| Former AZ-204 candidate | Data management, containers | Reusing outdated notes on dropped topics |
| Python ML engineer | Containers, messaging, security and monitoring | Overweighting model knowledge the exam never asks for |
| Database developer | Cosmos DB, containers, connecting services | Polishing familiar database skills while three domains go thin |
Which AI-200 profile fits before you book
Run three tests. Explain in one sentence each when you would pick Cosmos DB, PostgreSQL or Redis for vector search. Read a KEDA scale rule and a Kubernetes manifest and say what each does. Write a KQL query that joins on operation ID.
The one you cannot do names your starting profile. Confirm using the exam format, then the sample questions and a timed practice test. If you came from the old exam, AI-200 versus AZ-204 shows what was dropped. For the model side of the same builds, the AI-103 hub covers the apps and agents exam.