AI-200 prerequisites: what you need first
AI-200 lists no prerequisite exam in the Microsoft study guide I checked, but it assumes you already build software. The audience profile says you should be proficient in Azure SDKs, data management services, monitoring and troubleshooting, messaging and eventing, vector databases, Python programming and containerised applications. If several of those are new, plan more than the four weeks in the study plan.
What Microsoft says
The audience profile describes a candidate “responsible for contributing to all phases of implementing AI solutions on Azure, with an emphasis on back-end services and components”, and for supporting the whole development lifecycle: requirements, design, development, deployment, security and monitoring. The seven proficiencies above are the whole list.
No earlier certification, number of years or machine learning knowledge is named. Notably, the study guide’s “get trained” link still points at the old AZ-204 page, a sign that the two exams share a developer audience, but that is not a requirement to have held AZ-204.
The background that helps
Python
The exam shows code snippets and asks what they do or what is missing, so you need to read Python quickly: SDK clients, async calls, JSON, exceptions and environment configuration. You do not write programs from scratch in the exam.
Containers
Know what an image is, how a registry stores it, and why a container is not a virtual machine. If you have written a Dockerfile and run a container locally, you have the foundation for Container Registry, Container Apps and AKS. A Kubernetes manifest should look readable, even if you never wrote one.
Databases, including vectors
Cosmos DB for NoSQL, Azure Database for PostgreSQL and Azure Managed Redis all appear. Useful prior knowledge: partition keys, indexes, query cost, caching with expiry, and the idea of an embedding, a list of numbers that lets you search by meaning. See data management explained.
Messaging and events
Understand the difference between a message you must process (a queue) and an event that merely announces something happened. Retries and dead-letter handling are easier if you have used any queue before.
Security and monitoring basics
Secrets belong in a vault, not in code. For monitoring, the exam names OpenTelemetry tracing and KQL queries, so write a few log queries beforehand if you never have.
What you do not need
| Not needed | Why |
|---|---|
| Machine learning or data science | You build the infrastructure around models, not the models |
| Model or prompt tuning | Not an objective on the exam |
| AZ-204 | Retired; not required for AI-200 |
| A Kubernetes administrator role | Manifests and troubleshooting, not cluster operation |
| Another certification | None is listed as required |
Self-check
You are ready to begin if you can:
- Write a Python snippet that connects to a service with an SDK and handles an error.
- Explain the difference between an image and a running container.
- Say what a partition key is for.
- Explain why a queue and an event notification are different.
- Describe where an application should read a database password from.
Two or more gaps? Build one small project first: a Python API in a container that stores a record in a database.
If you come from a different background
Developers who know Azure but not AI will find the vector search parts new but small. Data scientists will be stronger on embeddings and weaker on containers and messaging; is AI-200 hard? shows how the difficulty shifts. If you have no development experience, start with AI-901 and build projects first.
Where it leads
Read the exam format, then AI-200 versus AZ-204 if you studied the old exam. Then follow the study plan. For the model side of the same stack, see AI-103; for production machine learning, AI-300.