Microsoft · AI-200
Microsoft Certified: Azure AI Cloud Developer Associate (AI-200) Practice Test
Validates the ability to design, build, and implement AI solutions on Azure with an emphasis on back-end services, scalable architectures, containerized workloads, and the full development lifecycle. Covers Azure SDKs, data management services, messaging and eventing, vector databases, and AI solution monitoring.
Exam Details
Questions
600
Duration
120 minutes
Passing Score
700/1000
Difficulty
AssociateLast Updated
Jun 2026
Topics Covered
Develop containerized solutions on AzureDevelop AI solutions using Azure data management servicesConnect to and consume Azure servicesSecure, monitor, and troubleshoot Azure solutionsAzure messaging and eventingVector databases and semantic retrievalPython programming and Azure SDKs
Microsoft Certified: Azure AI Cloud Developer Associate (AI-200) Practice Exam Preparation
Use this AI-200 practice exam to prepare for Microsoft Certified: Azure AI Cloud Developer Associate (AI-200) with realistic questions, detailed explanations, and focused study modes. The practice bank includes 600 questions for Microsoft AI-200, so you can review the exam steadily instead of relying on one long cram session.
As you practice, pay extra attention to recurring topics such as Develop containerized solutions on Azure, Develop AI solutions using Azure data management services, Connect to and consume Azure services, Secure, monitor, and troubleshoot Azure solutions, and Azure messaging and eventing. Start with short sessions to identify weak areas, then move into timed quizzes once your accuracy is consistent.
The explanations are especially useful when you want to connect exam wording to the responsibilities and scenarios described in the official certification guidance. Use the free preview first, then unlock the full question bank when you are ready to build a complete study routine.
Exam Domain Breakdown
Develop containerized solutions on Azure23%
Develop AI solutions by using Azure data management services28%
Connect to and consume Azure services23%
Secure, monitor, and troubleshoot Azure solutions23%
Exam Overview
Microsoft Certified: Azure AI Cloud Developer Associate (AI-200) validates a developer's ability to design, build, and implement production-grade AI solutions on Azure with a strong emphasis on back-end services, scalable architectures, and the full development lifecycle. The certification covers a broad set of Azure capabilities including containerized workloads on Azure Container Registry, Azure Container Apps, and Azure Kubernetes Service (AKS); AI-ready data management using Azure Cosmos DB for NoSQL and Azure Database for PostgreSQL with pgvector; event-driven integration through Azure Service Bus and Azure Event Grid; and serverless compute via Azure Functions. It also addresses operational concerns such as secret management with Azure Key Vault, distributed tracing with OpenTelemetry, and log analytics using KQL.
Launched in beta in May 2026 and targeting general availability in July 2026, AI-200 serves as the successor to the retiring AZ-204 exam. It shifts Microsoft's developer certification track toward the skills required to build and operate AI-integrated cloud systems, reflecting enterprise demand for back-end developers who can connect AI models to scalable, observable, and secure cloud infrastructure. Passing the exam also makes candidates eligible to pursue the Azure DevOps Engineer Expert certification.
Who Should Take This Exam
AI-200 is designed for back-end cloud developers who contribute to all phases of implementing AI solutions on Azure — from requirements gathering and design through deployment, security, and monitoring. Ideal candidates are software engineers or cloud developers with hands-on experience building server-side components and integrating Azure services into production systems, typically with 1–3 years of Azure development experience.
The certification is well-suited for developers transitioning from AZ-204, as well as those moving into AI-focused engineering roles where they must connect AI models to data pipelines, messaging systems, and containerized infrastructure. Job titles aligned with this certification include Azure Developer, Cloud Application Developer, AI Solutions Developer, and Backend Software Engineer working within AI product teams.
Prerequisites
Microsoft does not enforce formal prerequisites for AI-200, but candidates are expected to bring meaningful hands-on experience before sitting the exam. Proficiency in Python programming is assumed throughout, as is familiarity with Azure SDKs and common Azure services. Candidates should be comfortable working with containerized applications, including building and pushing container images, and should understand core concepts around REST APIs, cloud storage, and identity management.
A working knowledge of Azure data management services — particularly Azure Cosmos DB and Azure Database for PostgreSQL — is strongly recommended, along with exposure to vector databases and embedding-based retrieval patterns. Developers coming from an AZ-204 background will find many foundational topics carry over, though the AI-200 exam introduces additional depth around vector similarity search, RAG (Retrieval-Augmented Generation) patterns, semantic retrieval, and AI-specific observability tooling.
Exam Format
Exam AI-200 is a proctored assessment delivered online through Pearson VUE, available from a candidate's home or office. Candidates are allotted 120 minutes to complete the exam, and a score of 700 out of 1000 is required to pass. The exam is currently in beta, which means scores are not released immediately — rescoring begins at general availability launch and final results are typically delivered approximately 10 days after the live exam launches.
The exam may include interactive components in addition to standard question types (multiple choice, case studies, drag-and-drop, etc.), which candidates can preview in Microsoft's exam sandbox at https://go.microsoft.com/fwlink/?linkid=2226877. The exam is currently offered in English only, though additional languages may be added post-beta. Microsoft certifications earned at the Associate level expire annually and can be renewed for free via an online assessment on Microsoft Learn. A Practice Assessment is expected to become available within 8 weeks of the exam exiting beta.
Skills Measured
- 1.Develop containerized solutions on Azure (20–25%): Covers building, storing, versioning, and managing container images using Azure Container Registry (ACR) and ACR Tasks; deploying containers to Azure App Service with environment variable and secret configuration; deploying applications to Azure Container Apps including revision management and event-driven autoscaling via KEDA; deploying and managing workloads on Azure Kubernetes Service (AKS) using manifest files; and monitoring/troubleshooting containerized solutions by inspecting logs, events, and end-to-end connectivity.
- 2.Develop AI solutions by using Azure data management services (25–30%): Covers connecting to and querying Azure Cosmos DB for NoSQL via SDK; optimizing query performance using indexing policies, consistency levels, and Request Units; storing and retrieving embeddings and executing vector similarity search for semantic retrieval; implementing change feed processors; connecting to and querying Azure Database for PostgreSQL via SDK; designing schemas and indexing strategies including pgvector optimization; running vector similarity search and implementing RAG patterns with metadata filtering; configuring compute/memory/storage for vector workloads; and integrating Azure Managed Redis for caching, expiration, invalidation, and vector indexing.
- 3.Connect to and consume Azure services (20–25%): Covers building event- and message-based AI solutions using Azure Service Bus (including dead-letter queue handling, topics, and subscriptions) and Azure Event Grid (including filters, custom events, and retries); building serverless APIs using Azure Functions with triggers and bindings; and configuring and deploying function apps.
- 4.Secure, monitor, and troubleshoot Azure solutions (20–25%): Covers securing secrets using Azure Key Vault including secret rotation and retrieval; storing and retrieving application configuration using Azure App Configuration; tracing distributed systems using OpenTelemetry SDKs; and writing KQL (Kusto Query Language) queries to analyze logs and metrics.
Study Tips
- Download and work through the official AI-200 Study Guide at https://learn.microsoft.com/credentials/certifications/resources/study-guides/ai-200 — it lists every assessable skill with sub-bullets that reveal the exact depth of knowledge required for each topic.
- Build hands-on labs around the four exam domains: deploy a containerized Python app to Azure Container Apps with KEDA scaling, store and query embeddings in both Cosmos DB for NoSQL and PostgreSQL with pgvector, then wire them together with Service Bus and monitor with OpenTelemetry.
- Practice writing KQL queries in the Azure Monitor Logs interface. The exam tests real query authoring, so use the Azure documentation's KQL tutorial and the Log Analytics demo environment to get comfortable filtering, aggregating, and joining log tables.
- Focus significant study time on vector database patterns — specifically how to implement RAG (Retrieval-Augmented Generation) using pgvector in Azure Database for PostgreSQL and the native vector search in Cosmos DB for NoSQL. These are new areas absent from AZ-204 and carry heavy exam weight within the 25–30% data management domain.
- Use the official exam sandbox (https://go.microsoft.com/fwlink/?linkid=2226877) to familiarize yourself with the interactive question types before exam day, as AI-200 may include drag-and-drop or lab-style components in addition to multiple-choice questions.
- Review the Azure SDKs for Python (azure-cosmos, psycopg2/asyncpg for PostgreSQL, azure-servicebus, azure-eventgrid, azure-keyvault-secrets, azure-functions) and practice writing code that uses each. The exam is code-oriented and expects you to identify correct SDK usage patterns.
- Study Azure Container Registry Tasks and AKS manifest-based deployments specifically — the exam distinguishes between App Service, Container Apps, and AKS deployment models and expects you to know when and how to use each, including KEDA-based autoscaling configuration for Container Apps.
Career Benefits
AI-200 positions developers at the intersection of cloud engineering and applied AI — one of the fastest-growing specializations in enterprise technology. As the designated successor to AZ-204, it signals to employers that a candidate can not only build cloud-native applications but can architect the back-end infrastructure that powers production AI systems, including vector retrieval pipelines, event-driven AI workflows, and observable, containerized deployments. Roles commonly pursued by AI-200 holders include Azure AI Developer, Cloud Solutions Engineer, AI Platform Engineer, and Backend Developer on AI product teams, with Azure developer salaries in the United States ranging from approximately $110,000 to $160,000 annually depending on seniority and location.
Beyond job titles, AI-200 also serves as a prerequisite for the Azure DevOps Engineer Expert certification, making it a strategic stepping stone for developers targeting senior or lead engineering credentials. As enterprises accelerate AI adoption on Azure, demand for developers who can build reliable, secure, and scalable AI backends — rather than just call AI APIs — is increasing sharply, giving this certification strong near-term market relevance.
Sample Questions
5 sample questions with answers and explanations. Start a practice session to test yourself across all 600 questions.
Preview — answers shown1. Litware AI Processing needs to scale its image analysis containers in Azure Container Apps based on Azure Service Bus queue depth. The solution must scale to zero replicas when the queue is empty and allow up to 20 replicas during peak load. Which two KEDA scale rule configuration properties must the team set to define these scaling boundaries? (Select two!)
Multiple correct answersAminReplicas set to 0 to enable scale-to-zero when no messages are queued
BminReplicas set to 1 to maintain at least one warm replica at all times
CmaxReplicas set to 20 to cap the maximum number of replicas during peak load
DreplicaCount set to 20 as the desired steady-state instance count
EpollingInterval set to 30 seconds to control the frequency of queue depth checks
Explanation
Setting minReplicas to 0 is the required configuration to enable scale-to-zero behavior in Azure Container Apps KEDA integration, allowing all replicas to terminate when no messages are present in the Service Bus queue. Setting maxReplicas to 20 defines the upper bound on concurrent replicas during peak processing. These two properties directly control the scaling boundaries for the KEDA-driven Container App. Setting minReplicas to 1 prevents scale-to-zero by maintaining one active replica regardless of queue depth. replicaCount is not a valid KEDA scaling boundary property in Container Apps. pollingInterval controls how frequently the scaler checks the event source but does not define the minimum or maximum replica count.
2. Alpine Ski House is building an AI-powered booking notification platform. When a customer makes a reservation, a message is published to an Azure Service Bus topic named bookings. Three downstream AI services must each process the same message independently: a personalization service that processes only messages where the region property equals Pacific Northwest, an analytics service that must receive all booking messages, and a fraud detection service that processes only messages where the totalAmount property exceeds 5000. Which Service Bus configuration correctly implements this routing architecture? (Select one!)
ACreate three separate topics named bookings-personalization, bookings-analytics, and bookings-fraud and update the publisher to send each message to all three topics
BCreate a single queue named bookings with three competing consumers and implement region and totalAmount filtering logic inside each consumer's message handler
CCreate three subscriptions on the bookings topic and configure SqlFilter rules on the personalization and fraud detection subscriptions with a TrueFilter on the analytics subscription
DCreate one subscription with three consumer groups and assign a FilterExpression property to each consumer group in the Azure portal
Explanation
Azure Service Bus topics with multiple subscriptions are the correct pattern for publish-subscribe routing where multiple consumers must independently process the same message. Each subscription receives its own copy of every message published to the topic. SqlFilter rules use SQL-like expressions to filter which messages each subscription receives — the personalization subscription uses region = 'Pacific Northwest', the fraud detection subscription uses totalAmount > 5000, and the analytics subscription uses a TrueFilter that matches all messages without filtering. A single queue uses competing consumer semantics where each message is processed by exactly one consumer, preventing multiple services from independently processing the same booking. Creating three separate topics would require the publisher to send to three different endpoints, introducing coupling and operational complexity that the topic-subscription model avoids. Consumer groups are an Azure Event Hubs concept and do not exist in Azure Service Bus; Service Bus uses subscriptions to achieve per-consumer message copies.
3. Relecloud is building an AI customer support chatbot for their e-commerce platform. The chatbot must classify customer intents such as track order, process refund, and update address while extracting standard entities like order IDs and addresses from natural language messages. It must also extract domain-specific entity types unique to Relecloud — including internal product catalog codes and supplier IDs — that do not exist in any pre-built language model. Which two Azure AI Language capabilities should the developer configure? (Select two!)
Multiple correct answersAConversational Language Understanding (CLU) to train intent classifiers and entity extractors on Relecloud's labeled support conversation data
BSentiment Analysis to measure the emotional tone of customer messages and route frustrated customers to human agents
CCustom Named Entity Recognition (NER) to train a domain-specific model that recognizes Relecloud's internal product catalog codes and supplier IDs
DKey Phrase Extraction to identify the most important words and phrases from each customer message
EEntity Linking to connect recognized text spans in customer messages to entries in the Wikipedia knowledge base
Explanation
Conversational Language Understanding (CLU) is Microsoft's current service for building conversational AI with intent classification and entity extraction — it is the direct replacement for LUIS and supports training on custom intents and standard entity types from labeled conversational data. Custom Named Entity Recognition extends Azure AI Language to extract domain-specific entity types that are absent from pre-built models, such as proprietary catalog codes and supplier identifiers, by training on labeled examples. Both services can be deployed within the same Azure AI Language resource. Sentiment Analysis measures positive, negative, or neutral emotional tone but cannot classify business intents or extract structured typed entities from messages. Key Phrase Extraction surfaces prominent terms from free text but does not produce intent classifications or typed entity spans required for routing and data extraction. Entity Linking maps recognized entity mentions to a Wikipedia knowledge base, which provides no value for extracting proprietary internal identifiers that have no public knowledge base entries.
4. Litware's security operations team must investigate whether applications accessed Azure AI service API keys stored in Key Vault during the past 30 days. The team confirms that Key Vault diagnostic settings were configured with the Resource specific destination mode when routing logs to the Log Analytics workspace. Which table and filter should the team use in their KQL query to identify secret retrieval events? (Select one!)
AQuery AZKVAuditLogs and filter where OperationName == "SecretGet"
BQuery AZKVAuditLogs and filter where OperationName == "VaultGet"
CQuery AzureDiagnostics and filter where OperationName == "SecretGet" and Category == "AuditEvent"
DQuery AzureDiagnostics and filter where OperationName == "SecretRead"
Explanation
When Key Vault diagnostic settings are configured with the Resource specific destination mode, audit events are written to the dedicated AZKVAuditLogs table in the Log Analytics workspace. This table provides a strongly typed, Key Vault-specific schema with purpose-built columns for vault operations, caller identity, result codes, and secret identifiers. The AzureDiagnostics table is populated instead when the legacy Azure diagnostics mode is selected — it is a single flat table shared across many Azure resource types and is the correct target only for that older collection mode, not resource-specific mode. The correct operation name for reading a secret value is SecretGet, which is recorded every time a caller retrieves the value of a secret from the vault. VaultGet is the operation name logged when vault-level metadata such as SKU, access policies, and network rules is retrieved; it does not capture individual secret access events. SecretRead is not a valid Key Vault operation name — the naming convention follows an ObjectVerb pattern (SecretGet, SecretSet, SecretDelete, KeySign, VaultGet), and SecretRead does not exist in the audit schema.
5. Best For You Organics runs an AI-powered recipe approval workflow. Whenever a recipe is approved, three independent downstream services must each receive and process the event: a label printing service, an inventory forecasting service, and a customer notification service. Each service must process events at its own pace and cannot block or interfere with the other two. Which Azure Service Bus messaging pattern should the developer implement? (Select one!)
AA Service Bus queue with message sessions enabled and one session identifier assigned per consuming service
BThree separate Service Bus queues with the approval workflow duplicating each event into all three queues before returning
CA single Service Bus queue shared by all three services using the competing consumers pattern
DA Service Bus topic with three subscriptions, one dedicated subscription per downstream service
Explanation
A Service Bus topic with one subscription per downstream service is the correct fan-out pattern when multiple independent consumers must each receive a full copy of every message. A topic accepts published messages and delivers an independent copy to each subscription, where each subscription acts as its own virtual queue. This allows the label printing service, inventory forecasting service, and customer notification service to each consume events at their own pace without interfering with one another. A single shared queue uses the competing consumers pattern, where each message is delivered to exactly one consumer — only one of the three services would receive each event rather than all three. Manually duplicating messages into three separate queues tightly couples the publisher to the consumer topology and introduces reliability risks if any enqueue operation fails. Message sessions on a queue provide ordered processing of related messages for a single consumer group and do not replicate messages to multiple independent receivers.
$7.99
One-time access to this exam
Full access to all 600 questions
Or $15/mo for all 253 exams
Detailed explanations
Free preview stays available