AI-200 vs AZ-204: what changed?
AI-200 replaced AZ-204 when the Azure Developer Associate certification retired on July 31, 2026. About half of what AZ-204 tested carries over — Functions, Service Bus, Event Grid, Key Vault, App Configuration, containers — but the storage domain has been rebuilt around vector databases, and Python is now the named language. If you studied for AZ-204, you are not starting from zero, but you are not close to done either.
The two exams side by side
| AZ-204 (retired) | AI-200 | |
|---|---|---|
| Certification | Azure Developer Associate | Azure AI Cloud Developer Associate |
| Status | Retired July 31, 2026 | Current |
| Passing score | 700 | 700 |
| Named language | None in the audience profile | Python |
| Languages offered | 10 | 13 |
Domains compared
AZ-204 had five domains. AI-200 has four.
| AZ-204 domain | Where it went in AI-200 |
|---|---|
| Develop Azure compute solutions | Split: containers became their own domain; Functions moved to Connect to and consume Azure services |
| Develop for Azure storage | Replaced by Develop AI solutions by using Azure data management services |
| Implement Azure security | Merged into Secure, monitor, and troubleshoot Azure solutions |
| Monitor, troubleshoot, and optimize Azure solutions | Merged into the same domain |
| Connect to and consume Azure services and third-party services | Kept, narrowed to messaging, eventing and Functions |
What carries over
If you prepared for AZ-204, these parts of your knowledge transfer almost directly:
- Azure Container Registry and deploying containers to App Service.
- Azure Functions: triggers, bindings, and configuring and deploying function apps.
- Service Bus queues, topics and subscriptions, and Event Grid events and filters.
- Key Vault for secrets and App Configuration for settings.
- Cosmos DB for NoSQL: SDK access, consistency levels and the change feed.
Refresh them against the new outline rather than relearning them.
What is new
This is where AZ-204 material will not help you.
Vector search in three services. Cosmos DB for NoSQL gets vector embeddings and similarity search. Azure Database for PostgreSQL gets its own section with pgvector, indexing strategy, compute sizing for vector workloads and retrieval-augmented generation with metadata filters. Azure Managed Redis appears with caching and vector indexing. Nothing like this was on AZ-204.
Container orchestration. AZ-204 touched Container Apps and Container Instances. AI-200 goes further: revision management and KEDA scaling in Container Apps, deploying to Azure Kubernetes Service with manifest files, and troubleshooting both through logs and events.
ACR Tasks. Building images in the registry rather than on your own machine is now an explicit objective.
OpenTelemetry and KQL. Monitoring is framed around tracing distributed systems with OpenTelemetry SDKs and writing KQL queries against logs and metrics, rather than around Application Insights features.
Dead-letter handling and retries. Service Bus dead-letter queues and Event Grid retries are named explicitly, which pushes questions towards failure scenarios rather than the happy path.
What was dropped
Several AZ-204 topics do not appear in the AI-200 skills outline:
- Blob Storage development, including lifecycle management and metadata
- API Management
- Authentication with the Microsoft identity platform and Microsoft Graph, as standalone objectives
- Shared access signatures
- Azure Cache for Redis, replaced by Azure Managed Redis
Microsoft notes that related topics may still appear, so do not treat these as forbidden ground. But they are no longer where the marks are.
Is it a harder exam?
Different rather than harder. AZ-204 was broad and shallow across many services. AI-200 is narrower in service count but deeper in the data tier: you need to know why an IVFFlat index behaves differently from HNSW, or why an indexing policy affects RU cost, not just that the features exist. See is AI-200 hard? for the detail.
If you already hold AZ-204
Your credential remains visible on your Microsoft Learn profile, but it can no longer be renewed because the renewal assessment has been retired too. If the credential matters to your job, plan to take AI-200 before it lapses.
The shortest route is the four-week study plan, spending less time in week three (messaging and Functions) and more in week two (the data services).