AI-200 on One Page: The AZ-204 Replacement

If the reel sent you here: yes, the cert you planned for is gone. The good news is that about half of what you studied for AZ-204 still counts. This page is the lookup sheet. Keep the official AI-200 study guide open beside it, because Microsoft revises the outline.

📋 The exam at a glance

Try the exam sandbox once before booking so the interface is not new on the day.

⚖️ Skill areas and weights

Weights from the official study guide, skills measured as of October 7, 2026.

The data area is both the heaviest and the newest. That is where most AZ-204 holders need the extra hours.

🔄 What changed from AZ-204

Dropped

Kept

New

Already hold AZ-204?

• It stays valid until its current expiry date.
• It cannot be renewed. The renewal assessment retired with the exam.
• After expiry it shows on your transcript as expired.
• No upgrade path. To hold the new cert, you pass AI-200.

🧰 Services that matter, one thing each

The comparison that shows up again and again is Service Bus versus Event Grid. Service Bus carries work. Event Grid carries news.

🧭 Vector search in Cosmos DB for NoSQL

This is the newest material for most people, so here is only what the outline asks for: store embeddings, index them, query them. Learn the same three verbs for pgvector and Managed Redis too, and know when you would pick each.

Step 1: turn the feature on

Vector search is an account capability. It is auto-approved but can take about 15 minutes to apply.

az cosmosdb update \
  --resource-group rg-ai200 \
  --name cosmos-ai200 \
  --capabilities EnableNoSQLVectorSearch

Step 2: the container vector policy

The policy tells Cosmos DB where the vectors live and how to compare them.

{
  "vectorEmbeddings": [
    {
      "path": "/contentVector",
      "dataType": "float32",
      "distanceFunction": "cosine",
      "dimensions": 1536
    }
  ]
}

Step 3: pick the vector index type

quantizedFlat and diskANN need at least 1,000 vectors before they kick in. Below that, Cosmos DB runs a full scan. Exclude the vector path from the regular index so it is not indexed twice:

{
  "indexingMode": "consistent",
  "automatic": true,
  "includedPaths": [ { "path": "/*" } ],
  "excludedPaths": [
    { "path": "/_etag/?" },
    { "path": "/contentVector/*" }
  ],
  "vectorIndexes": [
    { "path": "/contentVector", "type": "diskANN" }
  ]
}

Step 4: query with VectorDistance

SELECT TOP 10 c.title, VectorDistance(c.contentVector, @queryVector) AS SimilarityScore
FROM c
ORDER BY VectorDistance(c.contentVector, @queryVector)

⏭️ What to skip

These were AZ-204 staples. They are not in the AI-200 outline, so do not spend hours on them.

• Microsoft identity platform, MSAL sign-in flows and Microsoft Graph
• Shared access signatures
• Blob Storage tiers, lifecycle and SDK
• API Management policies
• Event Hubs and Queue Storage
• Container Instances
• App Service deployment slots, autoscale and TLS
• Model training, tuning or evaluation. That is the AI engineer and data scientist track, not this one

Two things are not on the outline but are assumed knowledge: managed identities and basic Application Insights. You do not need to study them as topics, but you need to be able to use them.

🗓️ Four-week study plan

One week per skill area, with the heaviest one first while your energy is highest. If you already passed AZ-204, you can compress weeks 3 and 4 into one and call it a three-week plan.

Write the Python yourself every week. Reading SDK samples is not the same as connecting to Cosmos DB, sending a Service Bus message and reading a Key Vault secret from your own script.

🧪 Where to practise

Microsoft Learn

CloudLearn labs

Real Azure environment, guided steps, automatic checks, nothing left running afterwards.

For the assumed skills: Securing Azure Web Apps with Managed Identity and Key Vault and Getting Started with Azure Application Insights.

If you only do one lab this week, make it the KEDA one. Watching a container app scale from zero when messages hit the queue teaches more than any page about scale rules.

Hands-on LabCloudlearn.io

Container Apps KEDA Scaling with Service Bus Queue Workers

Wire a KEDA scale rule to a Service Bus queue and watch the app scale in a real Azure environment. Two AI-200 skill areas in one lab.

labs.cloudlearn.ioStart the lab

✅ Exam-day checklist

• Every outline item ticked, with something built for each
• Practice assessment taken at least once
• Exam sandbox tried
• Time limit per question set, slow ones marked for review
• Marked questions answered before taking the 5-minute break, because you cannot go back after it
• For online exams: clear desk, everything closed, Pearson VUE rules read the day before
• Booked at the time of day you think best

Where to go next after AI-200, and how it fits the rest of the 2027 map, is in the Azure Certification Roadmap 2027.