AZ-204 Exam Retired: The AI-200 Guide for Indian Azure Developers

You can no longer take AZ-204. Microsoft retired the exam, the Azure Developer Associate certification and its renewal assessment on 31 July 2026. The replacement is AI-200, Developing AI Cloud Solutions on Azure, which earns Microsoft Certified: Azure AI Cloud Developer Associate. It keeps containers, Azure Functions, messaging and Key Vault, and adds vector search, AKS and Python.

At a glance

• AZ-204 status: exam, certification and renewal assessment retired on 31 July 2026
• Replacement exam: AI-200, Developing AI Cloud Solutions on Azure, available from 9 July 2026
• Certification: Azure AI Cloud Developer Associate, intermediate level
• Skills measured: as of 7 October 2026, in four areas
• Language: Python is named in the audience profile
• Pass mark: 700 or greater
• Exam time: 100 minutes for role-based exams, or 120 if the exam contains labs
• Renewal: free online assessment every year

What the AZ-204 retirement means for you

If you already hold Azure Developer Associate, Microsoft's credential retirement page says a certification earned or renewed before retirement stays in the Active section of your transcript until it expires, then moves to Historical. The renewal assessment is retired, so you cannot extend it. Plan to pass AI-200 before your expiry date if you want an active developer credential on your profile without a gap.

If you were halfway through AZ-204 preparation, very little is wasted. Container Registry, Container Apps, App Service, Functions, Cosmos DB, Service Bus, Event Grid and Key Vault all carry over. What changes is the emphasis: AI-200 cares far more about data services built for AI workloads.

If you work for a Microsoft partner, which many Indian IT services and consulting firms are, Microsoft's July 2026 Partner Center announcements list Azure AI Cloud Developer Associate as the replacement for Azure Developer Associate in the AI Apps specialisation and the Digital and App Innovation designation. Certifications earned before retirement keep counting for partner requirements for one year after the retirement date. That is why your delivery manager may now be asking for AI-200 rather than AZ-204.

If AZ-204 was your route to DevOps Engineer Expert, check the prerequisite list on the DevOps Engineer Expert certification page before you book anything. Our AZ-400 guide for India covers that path.

AZ-204 versus AI-200: what moved

The table compares the last AZ-204 skills outline (January 2026) with the AI-200 outline of 7 October 2026.

One detail to note: the AI-200 study guide's skills-at-a-glance list gives data services 30 to 35%, while one section heading in the same guide still says 25 to 30%. The change log says that area's share increased, so plan for the higher figure.

No longer named in the skills measured: Blob Storage operations and lifecycle policies, API Management, Event Hubs, Queue Storage, Microsoft Graph, shared access signatures, user sign-in with the Microsoft identity platform, deployment slots and App Service autoscaling.

New: AKS deployments with manifest files, KEDA scaling in Container Apps, ACR Tasks, vector similarity search in Cosmos DB and PostgreSQL, Redis vector indexing, dead-letter queue handling, OpenTelemetry tracing and writing KQL.

Language: the AZ-204 profile asked for proficiency with Azure SDKs without naming a language. The AI-200 profile names Python. Many developers in Indian enterprise teams work mainly in C# or Java. If that is you, budget time for Python before week one, not during it.

The four AI-200 skill areas, and what to build

The AI-200 study guide is the authority. Here is how to turn each area into lab work. Keep all lab resources in one Indian region, such as Central India in Pune, so they sit together and you can delete them in one go.

Develop containerized solutions on Azure (20 to 25%)

Build and version images in Azure Container Registry, use ACR Tasks to build in the cloud, deploy to App Service with environment variables and secrets, deploy to Container Apps with revisions and KEDA scaling, deploy to AKS with manifest files, and troubleshoot with logs and events.

Practise: build one small Python API image with az acr build, deploy it to Container Apps, add a scale rule driven by a Service Bus queue, then deploy the same image to AKS with a manifest. Break the AKS deployment on purpose (a wrong image tag, a missing secret) and fix it from the events and logs. That is exactly what the troubleshooting objective asks for.

Develop AI solutions by using Azure data management services (30 to 35%)

The largest area covers three services. In Cosmos DB for NoSQL: SDK queries, indexing policies, consistency levels, request unit tuning, embeddings with vector search, and the change feed processor. In Azure Database for PostgreSQL: schema and index design, pgvector tuning, compute sizing for vector workloads, RAG with metadata filters, and connection optimisation. In Azure Managed Redis: caching, expiry, invalidation and vector indexing.

A Cosmos DB vector query in Python follows the pattern in Microsoft's documentation:

results = container.query_items(
    query=(
        "SELECT TOP 5 c.title, VectorDistance(c.embedding, @embedding) AS score "
        "FROM c ORDER BY VectorDistance(c.embedding, @embedding)"
    ),
    parameters=[{"name": "@embedding", "value": query_embedding}],
    enable_cross_partition_query=True,
)
for item in results:
    print(item["title"], item["score"])

For PostgreSQL, Microsoft's pgvector guide notes that you first add vector to the azure.extensions allowlist, then create it in each database:

CREATE EXTENSION vector;

CREATE INDEX ON products USING hnsw (embedding vector_cosine_ops);

SELECT id, name
FROM products
WHERE category = 'footwear'
ORDER BY embedding <=> '[0.12, 0.08, 0.33]'
LIMIT 5;

That last query is RAG with a metadata filter in miniature. Build the same product search twice, once on Cosmos DB and once on PostgreSQL, using a catalogue you invent for an imaginary online store. Then put Redis in front of the slower one and measure the difference. Comparing the two databases teaches you more than either alone.

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

Service Bus queues, topics and subscriptions with dead-letter handling; Event Grid custom events with filters and retries; and Azure Functions as serverless APIs with triggers and bindings.

Practise with an order pipeline: an HTTP-triggered Function accepts an order, drops it on a Service Bus queue, and a worker in Container Apps processes it. Send a malformed order and follow it into the dead-letter queue. Then publish an "order shipped" custom event to Event Grid and filter it so only one subscriber receives it.

Secure, monitor, and troubleshoot Azure solutions (20 to 25%)

Key Vault secrets with rotation and retrieval, App Configuration for settings, distributed tracing with OpenTelemetry SDKs, and KQL over logs and metrics. Use managed identities rather than connection strings in every lab from day one; it costs nothing extra and the secure answer is usually the right one.

Instrument your order pipeline with OpenTelemetry, then answer questions about it in KQL. For a workspace-based Application Insights resource:

AppRequests
| where TimeGenerated > ago(1h)
| summarize failures = countif(Success == false), p95 = percentile(DurationMs, 95) by Name
| order by failures desc

An eight-week plan for working developers

The plan assumes weekday evenings in IST plus one longer weekend session, and that you already write some Python.

Set a budget alert on day one and delete each week's resource group on Sunday night. AKS and PostgreSQL left running are the usual reasons for a surprise bill.

Exam day, booking and renewal

• Practice assessment: the free official one is on AI Skills Navigator, linked from the certification page, and you must sign in to launch it.
• Microsoft Learn is available during the exam: role-based exams let you open learn.microsoft.com, but the timer keeps running, so use it to check a parameter, not to research.
• Book with a personal Microsoft account: Microsoft's exam pages strongly recommend this, because records tied to a work account can be lost when you leave the organisation. In an industry where people change employers often, this matters.
• Price: exam fees depend on the country where you sit the exam. Check the exam page, and see our guide to Microsoft exam costs in India.
• Renewal: once earned, renew free each year in the six months before expiry.

If agents and Microsoft Foundry interest you more than back-end services, compare this path with our AI-103 exam guide.

How CertTulen Academy can help

CertTulen Academy, a Microsoft Training Services Partner, runs the AI-200 Azure AI Cloud Developer Associate course as live, MCT-led online training with hands-on labs across containers, vector databases, messaging and monitoring. Professional certification courses include an exam voucher, mock practice tests and lab access; the course page lists the details and INR pricing. Development teams can book a private corporate cohort. Message us on WhatsApp at +91 93618 83583 or use our contact page for a batch that suits your time zone.