AI Wisdom

Microsoft associate level

Microsoft Certified: Azure AI Cloud Developer Associate

Code
AI-200
Length
120 minutes
Pass mark
700 out of 1,000
Price
Set by country, shown when you book
Valid for
12 months, renewed by a free online assessment
Study time, our estimate
50 to 70 hours

The successor to AZ-204, and despite the name, it is a backend exam. There is little model work here. It covers the plumbing that AI apps run on: containers, vector storage, messaging and observability. Backend developers will find it the most comfortable Azure AI exam.

Take it if

You build the services around the model: APIs, queues, vector stores and containers on Azure.

Skip it if

You want to prove prompt, RAG or agent design skill. AI-103 covers that.

Before you start

Check yourself against this list first. If something is missing, close that gap before you pay for an exam.

  • You write Python against Azure SDKsMicrosoft lists this as a core skill.
  • You have run containersBuild and run one image before you start.
  • You know basic SQL and NoSQLThe data domain is the largest.
All exam details
Format
Usually 40 to 60 questions, per Microsoft
Length
120 minutes
Pass mark
700 out of 1,000
Price
Set by country, shown when you book
Valid for
12 months, renewed by a free online assessment
Delivery
Pearson VUE, online or test center
Prerequisites
None formally. Python and Azure SDK experience expected.
If you fail
Wait 24 hours, then 14 days

What the exam covers

  • Develop containerized solutions on Azure20–25%
  • Develop AI solutions by using Azure data management services25–30%
  • Connect to and consume Azure services20–25%
  • Secure, monitor, and troubleshoot Azure solutions20–25%

What to focus on

  • Vector workloads on Cosmos DB for NoSQL, PostgreSQL with pgvector, and Azure Managed Redis. This is the largest domain.
  • Cosmos DB basics that still catch people out: RU cost, indexing policies, consistency levels, and the change feed.
  • Container Apps with KEDA scaling, revisions, and AKS deployments from manifests.
  • Service Bus against Event Grid, including dead-letter handling.
  • Key Vault, App Configuration, OpenTelemetry and KQL.
Everything on one mapEvery topic and key idea in this certification.
mindmap
  root((AI-200))
    Containers
      ACR Tasks
      KEDA
      Revisions
    Data services for AI
      Cosmos DB
      pgvector
      Managed Redis
    Messaging and functions
      Service Bus
      Event Grid
      Functions
    Secure and monitor
      Key Vault
      App Configuration
      OpenTelemetry and KQL

Lesson 1 of 4, 20–25% of the exam

Containers

AI backends mostly run in containers, so this is the base layer.

ACR Tasks
Build images inside the registry, with no Docker needed on your CI agent.
KEDA
Event-driven scaling in Container Apps, including down to zero.
Revisions
Container Apps keeps versions, which helps with rollouts.
KEDA scaling on a queueReplicas follow the work, down to zero.
stateDiagram-v2
  direction LR
  [*] --> Zero
  Zero --> Running: Messages arrive
  Running --> MoreReplicas: Queue grows
  MoreReplicas --> Running: Queue shrinks
  Running --> Zero: Queue is empty

Common trap

Setting a fixed replica count for spiky queue work.

Try it yourself

Deploy a queue worker that scales to zero.

Official docs: ACR TasksScaling in Container Apps

Lesson 2 of 4, 25–30% of the exam

Data services for AI

The largest domain, and it grew in the October 2026 update.

Cosmos DB
Tune indexing policies to control RU cost, and use the change feed to react to changes.
pgvector
Add an index for vector search, or Postgres compares every row.
Managed Redis
Use it for caching and vector similarity.
Where the vectors can liveThe exam covers all three.
flowchart TD
  A{"Where is your data already?"} -->|"Documents in NoSQL"| B["Cosmos DB vector search"]
  A -->|"Relational tables"| C["PostgreSQL with pgvector"]
  A -->|"Need very fast lookups"| D["Azure Managed Redis"]
  B --> E["Tune indexing policy and RUs"]
  C --> F["Add a vector index"]
  D --> G["Set expiry and invalidation"]

Common trap

Indexing every path in Cosmos DB, which raises the cost of each write.

Try it yourself

Run the same vector search on Cosmos DB and pgvector, and record cost and latency.

Lesson 3 of 4, 20–25% of the exam

Messaging and functions

AI work is slow, so you need queues and events.

Service Bus
Durable work queues, with a dead-letter queue for failures.
Event Grid
Routes events to subscribers, with filters and retries.
Functions
Serverless APIs built with triggers and bindings.
Queue or event?Ask what the message means.
flowchart TD
  A{"What is the message?"} -->|"Work that must be done"| B["Service Bus"]
  A -->|"Something happened"| C["Event Grid"]
  B --> D["Failures go to the dead-letter queue"]
  C --> E["Subscribers filter what they want"]

Common trap

Using Event Grid when work must be processed reliably from a queue.

Try it yourself

Send failing messages to the dead-letter queue, then replay them.

Official docs: Service Bus dead-letter queuesAzure Event Grid

Lesson 4 of 4, 20–25% of the exam

Secure and monitor

Every design question has a security or observability angle.

Key Vault
Stores secrets, with rotation and retrieval.
App Configuration
Stores settings outside your code.
OpenTelemetry and KQL
Trace across services, then query logs and metrics.
One traced requestA trace shows where the time went.
sequenceDiagram
  participant G as API
  participant S as Worker service
  participant D as Database
  G->>S: traceparent header
  S->>D: Same trace continues
  D-->>S: Result
  S-->>G: Response
  Note over G,D: Query the spans with KQL

Common trap

Scaling up before finding where the time goes.

Try it yourself

Trace one request through three services, and write a KQL query for its slowest step.

Official docs: Azure Event Grid

Study plan

  1. Weeks 1 and 2: data

    Store embeddings and run vector search on Cosmos DB and on pgvector. Compare index types, and measure RU cost per query.

  2. Week 3: containers

    Build an image with ACR Tasks, deploy it to Container Apps with KEDA scaling, then deploy it to AKS.

  3. Week 4: messaging and observability

    Process work through Service Bus with a dead-letter queue, react to Event Grid events, and trace it all with OpenTelemetry.

  4. Exam week

    Take the free practice assessment and the mock below. Practice writing KQL by hand.

Exam day and after

  • You can open Microsoft Learn during associate and expert exams, but the clock keeps running. It helps with one lookup and not with twenty, so use it for exact names and limits.
  • Most questions cover generally available features, though preview features can appear if they are widely used.
  • Try the free exam sandbox first so the interface is not a surprise on the day.
  • The certification page was last updated on October 7, 2026, so check older notes against the current study guide.
  • Microsoft's guide shows the data domain at 30–35% in its summary but 25–30% on the section heading. Its change log says the share increased, so we use 30–35%.
Test yourself8 practice questions, scored by exam topic.
Start the mock exam

Keep going

Exam facts checked on October 31, 2026 against AI-200 study guide. Lessons are our own explanations, so use the official docs as the final word, and confirm price and dates before booking.