Free Preparation Discussions
MENU
Microsoft AI-200 Exam Questions
AI-200, Microsoft's Developing AI Cloud Solutions on Azure exam, is new enough that most of what's floating around still targets AZ-204, the exam it replaces. That's the first thing worth checking before you trust any question bank, including this one: does it track Microsoft's actual AI-200 outline, or is it recycled AZ-204 material with the exam code swapped? The AI-200 exam questions below come from the practice bank our candidates use before test day, built directly from Microsoft's current study guide. If your exam is close, start practicing now and check the domain weightings further down to see where you stand. If you're still early, the topic map and Microsoft's revision history will show you exactly what's changed.
| Exam name | Developing AI Cloud Solutions on Azure |
| Exam code | AI-200 |
| Certification | Azure Developer Associate |
| Practice questions in our bank | 142 |
| Questions on the real exam | 0 |
| Time allowed | 120 minutes |
| Passing score | 700 / 1000 |
| Exam cost | USD 165 |
| Retake policy | 24h after 1st fail, 14 days thereafter, max 5/year |
| Certification validity | 1 year, free online renewal |
AI-200 Exam Practice Questions
Free samples below come from the same 142-question AI-200 bank our candidates use, ordered by how Microsoft weights each domain.
You need to deploy a batch embedding workload according to the planned application architecture.
What should you use?
Correct Answer: A
The batch embedding workload is planned to run in Azure Kubernetes Service (AKS), and Kubernetes workloads should be defined declaratively by using YAML manifest files. Microsoft states that a Kubernetes manifest defines the desired state of cluster resources, including Deployments, Jobs, Services, ConfigMaps, and other Kubernetes objects. These manifests can then be deployed by using kubectl apply -f <manifest>.yaml.
For this case, YAML is particularly appropriate because the technical requirements emphasize repeatable, version-controlled, automated deployments. A batch vector embedding regeneration process would typically be represented as a Kubernetes workload such as a Job or another batch-oriented resource, with its configuration committed to Git alongside the rest of the deployment definitions.
kubectl run and kubectl create are imperative commands and are less appropriate for a controlled Git-based deployment model because they do not inherently preserve the complete desired-state definition as source-controlled configuration. XML is not the standard Kubernetes manifest format. az aks commands are primarily used to provision and manage the AKS cluster itself rather than declaratively define application workloads running inside the cluster.
Therefore, the correct choice is YAML-formatted Kubernetes manifest files.
Study Guide references: Azure Kubernetes Service Kubernetes manifests; declarative deployments; kubectl apply; Git-based workload configuration.
An AI application uses a database. The database credential rotates every 30 days.
The application currently requires a manual update each time the credential rotates.
You need to ensure that the application always uses the latest secret version without manual updates.
Which two actions should you perform? Each correct answer presents part of the solution. Choose two.
NOTE: Each correct selection is worth one point.
Correct Answer: C, D
A customer support bot built with Azure AI Foundry Agent Service must ground its answers strictly in an internal knowledge base and avoid hallucinating unsupported facts.
What should you implement?
Correct Answer: B
Connect the agent to an Azure AI Search knowledge source and explicitly constrain its instructions to answer from retrieved evidence. Microsoft Foundry Agent Service can use Azure AI Search to retrieve proprietary organizational content and ground model responses in that content. Retrieved documents provide the factual context from which the model generates its answer, and Azure AI Search can return source references for attribution.
Microsoft's RAG guidance also specifically recommends using clear system instructions that require the model to stay within retrieved content when hallucination risk must be reduced. If the available evidence does not support an answer, the instructions should direct the agent to state that the information is unavailable rather than inventing facts. Retrieval grounding substantially reduces unsupported generation, although production systems should still evaluate outputs because grounding alone does not mathematically guarantee zero hallucinations.
Increasing temperature would generally increase output variability and is contrary to the objective. A larger context window without retrieval does not give the model access to the organization's internal knowledge base. Disabling content filtering weakens safety controls and has no role in factual grounding.
Study Guide references: Microsoft Foundry Agent Service Azure AI Search tool; RAG grounding; knowledge sources; source attribution; grounding instructions.
===============
An AI platform uses App Configuration for feature flags and endpoint routing.
The platform stores secrets alongside configuration data and does NOT support dynamic configuration refresh.
You need to redesign the configuration approach so that:
configuration supports dynamic refresh;
secrets are stored outside App Configuration;
the application can securely access both App Configuration and Key Vault.
You need to enable secure dynamic configuration management for the platform.
Which three actions should you perform? Each correct answer presents part of the solution. Choose three.
NOTE: Each correct selection is worth one point.
Correct Answer: D, E
You develop an ASP.NFT Core app that uses Azure App Configuration. You also create an App Configuration containing 100 settings. The app must meet the following requirements:
* Ensure the consistency of all configuration data when changes to individual settings occur.
* Handle configuration data changes dynamically without causing the application to restart.
* Reduce the overall number of requests made to App Configuration APIs
You must implement dynamic configuration updates in the app.
What are two ways to achieve this goal? Each correct answer presents part of the solution.
NOTE: Each correct selection is worth one point.
Correct Answer: C, E
AI-200 Exam Topics & Weightings (Microsoft's Current Outline)
Weightings below are pulled straight from Microsoft's official skills outline.
Develop containerized solutions on Azure
20–25%From container image to running workload, this domain is the build-and-ship half of the exam: Azure Container Registry and ACR Tasks for images, App Service deployment with the right environment variables and secrets, then orchestration, Container Apps with environment and revision management, KEDA-driven autoscaling, and AKS deployments through manifest files. Expect questions that ask you to troubleshoot a deployment by reading logs, events, and end-to-end connectivity.
Develop AI solutions by using Azure data management services
25–30%The largest domain, and the one most specific to this exam: connecting to and querying Cosmos DB for NoSQL and Azure Database for PostgreSQL through their SDKs, storing and retrieving vector embeddings for semantic search, and implementing retrieval-augmented generation patterns with metadata filtering. Also covers indexing and consistency-level trade-offs for query performance, plus Azure Managed Redis for caching and vector indexing. Questions here tend to be trade-off scenarios: which data service or indexing strategy fits a stated latency, cost, or accuracy requirement.
Connect to and consume Azure services
20–25%Two Azure services do the connecting work here: Service Bus for queues, topics, subscriptions, and dead-letter queue handling, and Event Grid for event-driven workflows with filters, custom events, and retries. Azure Functions rounds it out, building and deploying serverless APIs with triggers and bindings. Questions typically describe a back-end integration requirement and ask which messaging or eventing pattern satisfies it without losing messages or over-provisioning compute.
Secure, monitor, and troubleshoot Azure solutions
20–25%This domain is where the exam gets diagnostic. Azure Key Vault handles secrets, including rotation and retrieval, and Azure App Configuration handles settings, but the real testing ground is monitoring: tracing distributed systems with OpenTelemetry SDKs and writing KQL queries to analyze logs and metrics. You're often shown a log or trace output and asked what it reveals about a failure, rather than asked to define a monitoring concept.
Official Revisions by Microsoft
April 15, 2026: — Microsoft published the AI-200 study guide, introducing this exam as the replacement for AZ-204 with an expanded focus on AI data services, vector databases, and retrieval-augmented generation patterns.
May 5, 2026: — Documentation update to the study guide. No changes to domain weightings.
Source: Microsoft's official AI-200 study guide. AI-200 is a newly launched exam, so this is its full revision history to date. Our question bank is updated to match each revision as Microsoft publishes them.
What the AI-200 Exam Really Tests (and How)
Microsoft's outline says "build and deploy a containerized solution" in one line and "optimize query performance" a few lines later. Those two verbs point at genuinely different question types, and reading the outline that way, verb by verb, tells you more about exam day than the topic list alone. The four cards below break down which verb maps to which format.
Hands-on build and deployment skills
Tested via: hot area, drag-and-drop & command-completion questionsAnything phrased as "build," "deploy," or "configure" in Microsoft's outline, container images in Azure Container Registry, Container Apps revisions, AKS manifests, can turn up as a portal screenshot where you click the right setting, a drag-and-drop deployment sequence, or a CLI or SDK command with the missing parameters. Microsoft's audience profile expects Python and Azure SDK proficiency, so code-completion questions are fair game. Microsoft keeps an exam sandbox open for exactly this reason: meet each interaction style there rather than in the first ten minutes of a live sitting.
Choosing the right data service under constraints
Tested via: scenario-based multiple-choice questionsShort scenarios give you a requirement, a query latency target, an RU budget, a storage cost limit, and ask which data service, indexing strategy, or consistency level meets it: Cosmos DB versus PostgreSQL for a given access pattern, which consistency level avoids stale reads, how to reduce pgvector compute overhead. These reward knowing the decision boundaries between similar services, not their feature lists.
Multi-step reasoning across services
Tested via: case-study question setsA case study presents one company, its existing architecture, technical requirements, and planned changes, followed by a block of questions spanning containers, data services, messaging, and security at once. Implementing a full retrieval-augmented generation pattern, for instance, touches vector storage, retrieval, and the API layer serving it. Once a case study is behind you it stays behind you, which makes the decision of when to move on part of the test.
Troubleshooting and interpreting telemetry
Tested via: exhibit-based questionsMicrosoft's outline repeatedly asks you to monitor, trace, and troubleshoot: inspect AKS or Container Apps logs and events, interpret a KQL query result, trace a distributed request through OpenTelemetry. On the exam these become exhibits, a log excerpt, a trace waterfall, a query output, where you diagnose what failed or what a metric is telling you.
How to Pass the AI-200 Exam on Your First Attempt
Some candidates have six months of runway before their AI-200 date. Most don't. If you're in the second group, the good news is that this exam rewards focused practice more than it rewards time on the clock. What actually costs you marks is reading service documentation cover to cover instead of drilling the scenarios that show up on test day. Here's the shortest sequence we know that holds up when exam day arrives.
Start with a full set, before you've studied anything. Ignore the score it produces and read the per-domain split instead. Twenty minutes of practice tells you more about where you stand than a week of reading documentation.
When you miss one, chase down why the right answer wins. Take the answer you chose to Microsoft Learn and read what that service or setting actually does. AI-200 reuses the same decision points, Cosmos DB versus PostgreSQL for vector search, Service Bus versus Event Grid for a given trigger, across many questions, so one understood mistake fixes several future ones.
Weight your remaining study time the way Microsoft weights the exam. Data management services alone carries 25–30%, the single largest domain and the most exam-specific. A weak grasp of vector search or RAG patterns costs more here than anywhere else on the exam.
Run exactly one timed, full-length simulation before test day. You're testing pacing at this point, not knowledge: case studies and any interactive components eat time, and you can't return to them once you move on. Whatever the simulation exposes is your final review list.
What to Look for in AI-200 Practice Questions
Why Prefer Pass4Success Practice Material
Picking AI-200 practice material comes down to one real question: will it hold up against the exam Microsoft is actually running right now, not the one that existed six months ago? A retake costs USD 165 Microsoft's standard associate-level exam fee and since AI-200 is currently a beta exam, a failed attempt means waiting for the exam to go fully live before you can retake it at all, not just a short cooldown. So getting it right the first time is worth more than saving a few dollars on prep. Here's what separates material worth your time from material that isn't.
Current with the latest exam revision
AI-200 launched recently as Microsoft's replacement for AZ-204, and Microsoft will keep revising it as Azure's AI services evolve. Questions written for an outdated outline, or worse, still built for AZ-204, look plausible and teach you the wrong service.
✓ Ours: revisions happen on a set schedule built around how Microsoft examines this material in practice, with syllabus updates handled as an extra pass rather than the only trigger. When Microsoft does revise the AI-200 outline, affected questions are re-checked as well. Whatever date that work last landed on is printed in the exam details table, so the claim is checkable rather than asserted.
A way to try before you pay
Nobody needs to take a prep site's word for question quality, not when a free demo settles it in five minutes. That matters even more for a brand-new exam like AI-200, where outdated or recycled material shows up fastest before the market catches up.
✓ Ours: both the PDF and the practice test open in demo form with no payment step in front of them, and the samples further up this page are drawn from the same set.
Timed, exam-style practice, not just a question list
Knowing an answer and picking it correctly under a countdown timer are two different skills, and AI-200's case studies (plus any interactive components) are built to expose the gap between them.
✓ Ours: timed sittings run in the practice test while the PDF handles review away from a screen, and a single purchase covers both rather than splitting them.
Enough volume to expose your weak areas
Four domains, and one of them (data management services) alone carries 25–30% of the exam. Twenty sample questions won't surface a weak spot that specific; you need enough per domain for a real pattern to show up.
✓ Ours: 142 AI-200 questions across the exam's four domains.
A cost that makes sense against the retake fee
Compare the cost of prep material to the wrong number and it always looks expensive. Compare it to the cost of booking the exam twice, and the math changes.
✓ Ours: $69 Practice material Compare that to a single $165 retake plus 24 hours wait before rebook you exam, and the math favors getting it right the first time.
AI-200 Exam Discussion: What Recent Candidates Say
Below, candidates who have sat AI-200 compare notes: the topics that surprised them, where marks slipped away, and what they would study differently a second time. If you have taken the exam, add what you learned.
Ready to Pass AI-200 on Your First Attempt?
Full access to all 142 AI-200 questions, in both the PDF and the practice test.
Get Premium AccessFrequently Asked Questions
What skills are measured on the AI-200 exam?
AI-200 splits into four domains, not five like some of Microsoft's older associate exams: developing containerized solutions on Azure (20–25%), developing AI solutions with Azure data management services (25–30%), connecting to and consuming Azure services (20–25%), and securing, monitoring, and troubleshooting Azure solutions (20–25%). The weightings come from Microsoft's study guide, last updated May 5, 2026. Each of those domains is unpacked in the section above, with the subskills Microsoft lists underneath it.
What are the prerequisites for taking the AI-200 exam?
Technically none. Microsoft lets you book AI-200 without any other certification behind you, though it replaces AZ-204 in the developer track. In practice, the exam assumes hands-on experience with Azure SDKs, Python, containerized applications, and vector databases, and candidates without real development work on Azure tend to struggle with the scenario questions regardless of what the prerequisite page says.
How hard is the AI-200 exam?
Ask a candidate who studied AZ-204 material by mistake how hard it is. The exam leans on applied development decisions across containers, data services, and messaging rather than definitions, and its case studies mix multiple Azure services in one scenario. Regular hands-on experience building AI backends on Azure makes it demanding but manageable. Without that, the vector database and RAG-pattern questions turn into guesswork.
What types of questions are on the AI-200 exam?
Multiple choice and multi-select, drag-and-drop, and hot area questions where you click the right part of a screenshot make up the bulk of it, alongside case studies built around a single company scenario. Microsoft's exam policy also notes the exam may include interactive components, which typically means hands-on lab-style tasks. Microsoft's free exam sandbox lets you try the non-lab formats before test day.
What are the most common mistakes on the AI-200 exam?
Vector search trips up more candidates than anything else on this exam. Choosing Cosmos DB, PostgreSQL, or Redis without weighing the stated cost or latency constraint, missing dead-letter queue or retry configuration in messaging questions, and reading KQL query results too fast are the recurring point-losers. Many are reading errors as much as knowledge errors: scenario questions usually bury one deciding constraint in the requirements.
How do the AI-200 topics connect in a real project workflow?
Picture a real AI backend project and the four domains stop looking separate. You containerize a service and deploy it to Container Apps or AKS, wire it to a data layer like Cosmos DB or PostgreSQL for retrieval, connect it to the rest of the system through Service Bus or Event Grid, then secure secrets in Key Vault and monitor the result with OpenTelemetry and KQL. AI-200's case studies mirror that: they routinely need decisions from more than one domain in a single scenario.
What is an effective final-week review plan for AI-200?
Skip the full re-read. Drill your weakest domain instead, and for most candidates that's data management services, since it carries the largest single weighting. Run one full timed practice test around midweek, review every question you missed against Microsoft Learn documentation, then stop adding new material. The day before, do a light pass over SDK syntax and KQL query structure, nothing more.
Is AI-200 worth it in 2026?
For a developer building AI-integrated backends on Azure, this isn't really optional anymore. AI-200 replaces AZ-204 as Microsoft's core associate-level developer certification and adds the vector database and RAG-pattern skills AZ-204 never covered. The certification renews free each year through an online assessment on Microsoft Learn, so the exam fee is a one-time cost as long as you keep renewing.
Where does AI-200 fit in the Microsoft Azure certification path?
AI-200 sits at the Associate level in Microsoft's Azure certification path. It replaces AZ-204, which is being retired in favor of this AI-focused track. Passing it earns the Microsoft Certified: Azure AI Cloud Developer Associate certification, validating your ability to build, integrate, and monitor AI solutions on Azure. If you're currently on the AZ-204 path, switching to AI-200 is the recommended move.
Related Microsoft Azure Exams
Because AI-200 just launched, most of what's below covers its first review cycle rather than a long history. We recheck the AI-200 bank on a fixed schedule that runs independent of when Microsoft touches the syllabus, and we'll extend that into a full revision-tracking record here once Microsoft starts revising the outline. That date is recorded in the exam details table near the start of this page.