AI-200 Certification Guide 2026: Azure AI Cloud Developer Associate

Microsoft is retiring the AZ-204 exam and the Azure Developer Associate certification on July 31, 2026 — along with the renewal assessments, so there is no renewal path after that date. The replacement is AI-200: Developing AI Cloud Solutions on Azure, which earns a new credential: Microsoft Certified: Azure AI Cloud Developer Associate. The exam runs 120 minutes, requires a 700/1000 score, and tests four domains built around containers, vector-enabled databases, event-driven messaging, and observability — with Python and vector search assumed from the first bullet of the audience profile.

This guide covers what the AI-200 certification actually tests (from the official study guide), what changed from AZ-204, what the retirement means if you hold or were preparing for the old exam, and how to prepare while the SERP and question-bank ecosystem around this exam are still nearly empty.

What Is the AI-200 Certification?

AI-200 is Microsoft's intermediate exam for developers who build the back end of AI solutions on Azure — the containers, databases, message queues, and monitoring that sit underneath the model calls. Passing it earns the Azure AI Cloud Developer Associate certification, positioned as the successor to Azure Developer Associate.

The audience profile is blunt about what it expects: proficiency in Azure SDKs, data management services, messaging and eventing, monitoring, vector databases, Python programming, and containerized applications. That list is the thesis of the whole refresh. Microsoft no longer thinks the default Azure developer builds web apps with some storage attached — it thinks they build AI application backends where embeddings, similarity search, and event-driven pipelines are routine work.

One launch-window caveat: the official practice assessment doesn't exist yet. Microsoft publishes those within about eight weeks of an exam leaving beta, so if you sit AI-200 this summer you're working from the study guide and hands-on practice, not from a Microsoft-provided question set.

AZ-204 Retirement: What Happens to Your Credential

AZ-204 and the Azure Developer Associate certification retire on July 31, 2026, at 11:59 PM Central Standard Time. The warning on the certification page is explicit: "This certification, related exam, and renewal assessments will retire on July 31, 2026. You will no longer be able to earn or renew this certification after this date."

The renewal-assessment retirement is the part that catches people. Microsoft associate certifications normally renew annually through a free online assessment — but that assessment is being switched off too. If you hold Azure Developer Associate, it stays on your transcript until its individual expiration date, and then it lapses with no way to extend it. This is the same pattern Microsoft used when AI-102 retired in June and announced for AZ-500's retirement in August: the old credential doesn't convert, it just ends. The forward path is passing the new exam.

If you're mid-preparation for AZ-204 right now, here's my honest take: only book it if an employer or contract requires that specific credential this year. Passing AZ-204 in July buys you a certification that dead-ends in twelve months with no renewal option — you'll be studying for AI-200 in 2027 anyway. If you're more than a few weeks from exam-ready, redirect the effort now. Roughly two-thirds of AZ-204's material carries over, and the sections that don't (the vector database work in particular) are the ones worth learning regardless.

AI-200 Exam Domains and Weights

The official study guide (skills measured as of April 15, 2026) defines four domains with unusually flat weighting — no domain falls below 20%:

Containerized solutions (20–25%)

Azure Container Registry (including ACR Tasks for building images), deploying containers to App Service with environment variables and secrets, Azure Container Apps with environment configuration and revision management, event-driven scaling with KEDA, and deploying to AKS with manifest files. AKS was never on AZ-204 — its inclusion here, alongside log- and connectivity-based troubleshooting, moves the exam much closer to what platform teams actually run.

AI solutions with Azure data services (25–30%)

The largest domain, and the most interesting one. Three services, each with a vector angle:

• Azure Cosmos DB for NoSQL — SDK operations, RU optimization through indexing policies and consistency levels, storing embeddings and running vector similarity search, and change feed processors for reacting to new items.
• Azure Database for PostgreSQL — schema and index design, pgvector query tuning, sizing compute and memory for vector workloads, and implementing retrieval-augmented generation patterns with metadata filtering.
• Azure Managed Redis — caching with expiration and invalidation, plus vector indexing for similarity search.

The pgvector bullets read like they were lifted from a production RAG retrospective — "optimizing query latency and reducing pgvector compute overhead" is exactly the tuning work that eats a sprint the first time a semantic search feature meets real traffic. If you want the conceptual grounding on why you'd pick one vector store over another before drilling the Azure specifics, our vector database comparison covers the trade-offs.

Connect to and consume Azure services (20–25%)

Azure Service Bus (queues, topics, subscriptions, dead-letter handling), Azure Event Grid (filters, custom events, retries), and Azure Functions (triggers, bindings, configuration, deployment). This is the surviving core of AZ-204's integration domain, trimmed to the services that matter for asynchronous AI pipelines — Event Hubs and Queue Storage are gone.

Secure, monitor, troubleshoot (20–25%)

Key Vault secret management including rotation, Azure App Configuration, distributed tracing with OpenTelemetry SDKs, and KQL queries against logs and metrics. Note the modernization: AZ-204 tested Application Insights instrumentation and gave monitoring 5–10% of the exam; AI-200 tests vendor-neutral OpenTelemetry plus hands-on KQL and bundles security and observability into nearly a quarter of the score.

AI-200 vs AZ-204: What Actually Changed

The short version: AI-200 keeps AZ-204's back-end developer role but rebuilds the service list around AI application infrastructure. It is not an AI-modeling exam — there's no prompt engineering or model deployment here (that's AI-103's territory). It tests the data and compute layer AI apps run on.

What was cut is as telling as what was added. API Management, Blob Storage lifecycle management, Event Hubs, Queue Storage, App Service deployment slots, and the entire Microsoft identity development block (MSAL, Microsoft Graph, shared access signatures) are gone from the skills list. Identity work didn't disappear from the platform — it moved to the security-focused exams like SC-500, while AI-200 keeps only the developer-facing pieces: Key Vault and App Configuration.

This is the fourth Microsoft exam rebuilt around AI workloads in 2026, after AI-901, AI-103, and SC-500 — at this point it's a strategy, not a trend. The practical consequence for candidates: every AZ-204 study resource, course, and practice bank published before mid-2026 is aimed at an exam that stops existing on July 31.

How to Prepare for AI-200

Start from the official study guide and build every listed skill hands-on — with no official practice assessment until roughly eight weeks after GA, hands-on work is the only reliable signal of readiness. A reasonable sequence:

1. Containers first. Push an image through ACR Tasks, deploy it to Container Apps, configure a KEDA scaler against a Service Bus queue, then deploy the same workload to AKS with manifests. That one exercise chain touches most of domain one and the messaging half of domain three.
2. Build a small RAG feature three times — once on Cosmos DB vector search, once on PostgreSQL with pgvector, once with Managed Redis as the vector index. The exam's largest domain is the differences between these: consistency levels and RUs on Cosmos, index tuning on Postgres, expiration semantics on Redis.
3. Instrument everything. Wire OpenTelemetry tracing into the app from step one and practice writing KQL against its logs. Monitoring quietly grew from AZ-204's smallest domain into a quarter of this exam, and it's where I'd bet under-prepared candidates lose the most points.
4. Walk the exam sandbox before test day — AI-200 is flagged as potentially including interactive components, and Microsoft's newer interactive formats reward familiarity with the interface.

One honest observation: as of this writing, the study guide's own training links still point at AZ-204 learning paths — the AI-200-specific Microsoft Learn content is clearly still being assembled. Early candidates are working from documentation and practice, which cuts both ways: preparation is harder, but the certification is scarcer.

Where AI-200 Fits in Your Azure Path

Microsoft's 2026 Azure track now has a clear shape: AI-901 for fundamentals, then a fork at associate level — AI-103 if you build the AI application layer (agents, Foundry, model integration), AI-200 if you build the services layer underneath it (containers, data, messaging, observability). They're complements, not competitors; a small team shipping an AI product needs both skill sets, and plenty of engineers will end up holding both.

Every domain on this exam is learn-by-doing material — you can't read your way to pgvector tuning or KEDA scaling behavior. CloudaQube's hands-on AI labs put you in real cloud environments building the same RAG, container, and pipeline patterns AI-200 tests, which is a considerably cheaper way to make mistakes than your first proctored attempt at an exam with no practice assessment.