Can I still take AI-102 in India?

No. AI-102 retired worldwide on 30 June 2026, together with the Azure AI Engineer Associate certification and its renewal assessment. AI-103 is the exam to book now. If you already hold Azure AI Engineer Associate, it stays active until its expiry date, but you cannot renew it.

Is AI-103 harder than AI-102?

It is different rather than simply harder. AI-103 drops several older topics, such as custom vision training and question answering knowledge bases, and puts far more weight on generative AI, agents, evaluation and monitoring. If you already build with large language model APIs, AI-103 will feel closer to your daily work than AI-102 did.

Do I need AI-901 before AI-103?

No. Microsoft lists no prerequisite certification for AI-103. AI-901 helps if you are new to Azure and Microsoft Foundry, because it covers the same services at a basic level. If you already deploy models and write Python against Azure, you can go straight to AI-103.

Can I prepare for AI-103 using C# instead of Python?

Plan for Python. The AI-103 audience profile asks for experience developing apps by using Python, and the official course is designed for engineers familiar with Python. Foundry has C# SDKs as well, but since the profile names only Python, prepare with the Python samples.

How long should a working developer in India spend on AI-103?

If you already write Python and have used a large language model API, about eight weeks of evening study with one longer weekend lab is a realistic plan. Build one small project per week, covering grounding, agents, vision, speech and extraction, rather than only reading modules.

AI-103 Exam Guide for India: Azure AI Apps and Agents Developer

AI-103, Developing AI Apps and Agents on Azure, is the exam for the Microsoft Certified: Azure AI Apps and Agents Developer Associate certification. It replaced AI-102, which retired on 30 June 2026. Expect Python, Microsoft Foundry and agents: generative AI and agentic solutions alone make up 30 to 35% of the exam, and you need a score of 700 to pass.

At a glance

Exam: AI-103, Developing AI Apps and Agents on Azure
Certification: Azure AI Apps and Agents Developer Associate, intermediate level, renewed free every year
Replaces: AI-102, retired 30 June 2026
Skills measured: as of 16 April 2026, in five areas
Language: Python; Microsoft expects experience developing apps in it
Exam time: 100 minutes for role-based exams, or 120 if the exam contains labs
Prerequisite certification: none listed by Microsoft

What happened to AI-102, and what it means for you

Microsoft retired AI-102 and the Azure AI Engineer Associate certification on 30 June 2026, and its renewal assessment is retired too. Microsoft's June 2026 Partner Center announcements name AI-103 as the replacement, available from 19 June 2026.

If you already hold Azure AI Engineer Associate, Microsoft's credential retirement page says it stays in the Active section of your transcript until its expiry date, then moves to history. You cannot renew it. If your employer is a Microsoft partner, certifications earned before the retirement date keep counting toward its partner requirements for one year after retirement. After that, it will want AI-103.

The weightings show how far the job has moved:

Skill area	AI-102 (Dec 2025)	AI-103 (Apr 2026)
Plan and manage an AI solution	20 to 25%	25 to 30%
Generative AI	15 to 20%	30 to 35%, with agents
Agentic solutions	5 to 10%	Merged into the row above
Computer vision	10 to 15%	10 to 15%
Language and text	15 to 20%	10 to 15%
Knowledge mining and extraction	15 to 20%	10 to 15%

What dropped out: training and labelling custom vision models, intent and utterance language models, custom question answering knowledge bases, and C# as an expected language. What arrived: multi-agent orchestration, approval flows and oversight modes, image and video generation, prompt injection hidden in images, tracing and token analytics, and CI/CD for Foundry projects. If your notes still talk about knowledge base exports, they belong to the old exam.

The five skill areas and what to practise

The AI-103 study guide is the authority. Here is how to turn each area into lab work.

Plan and manage an Azure AI solution (25 to 30%)

This covers choosing models (large, small, multimodal or Foundry Tools), retrieval and indexing methods, memory and tool services for agents, deployment options, CI/CD, quotas and rate limits, cost, monitoring for drift and grounding quality, and security with managed identity, private networking, keyless credentials and role policies. Responsible AI sits here too: guardrails, evaluators, trace logging and oversight modes for agents.

Practise: deploy the same model twice with different deployment types and compare. Then switch your app from API keys to keyless authentication with DefaultAzureCredential and a role assignment. Expect scenario questions where the keyless answer is the secure one.

Implement generative AI and agentic solutions (30 to 35%)

The heaviest area: deploying models, retrieval-augmented generation (RAG), multistep workflows, evaluating for fabrications and safety, agents with retrieval, function calling and memory, multi-agent orchestration, safeguards with approval steps, and observability.

Practise with a small project you are allowed to use, such as your college's academic regulations or a sample HR leave policy. Start with an agent defined in code, following Microsoft's Foundry SDK quickstart:

from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient
from azure.ai.projects.models import PromptAgentDefinition

project = AIProjectClient(
    endpoint="https://<resource-name>.services.ai.azure.com/api/projects/<project-name>",
    credential=DefaultAzureCredential(),
)

project.agents.create_version(
    agent_name="leave-policy-agent",
    definition=PromptAgentDefinition(
        model="<your-deployment-name>",
        instructions="Answer questions about the leave policy. If you are not sure, say so.",
    ),
)

openai = project.get_openai_client(agent_name="leave-policy-agent")
conversation = openai.conversations.create()
reply = openai.responses.create(
    conversation=conversation.id,
    input="Can I carry forward unused earned leave?",
)
print(reply.output_text)

This agent has no knowledge yet. Next, attach a search tool over the policy documents, add one custom function (a mock leave-balance lookup), and put a human approval step before any action that would change data. That one project covers grounding, function calling, memory and approval flows.

Implement computer vision solutions (10 to 15%)

Image and video generation and editing, captions, accessible alt text, question answering grounded in images, Content Understanding for visual features, video analysis, and responsible AI for images, including indirect prompt injection through text embedded in a picture.

Practise: build an alt-text generator for product photos, the kind an Indian D2C brand would need for its online store. Then test it with an image that contains the printed words "ignore your instructions" and see whether your filters catch it.

Implement text analysis solutions (10 to 15%)

Entity, topic and summary extraction into structured JSON, sentiment and sensitive content detection, translation with Azure Translator or a language model, and speech: speech to text, text to speech, speech as an agent modality and speech translation.

This is where Indian learners can build something genuinely local. Azure Speech supports speech to text for Indian locales including en-IN, hi-IN, ta-IN, te-IN, bn-IN, mr-IN, kn-IN and ml-IN (see language and voice support). Azure Translator covers Hindi, Tamil, Telugu, Marathi and other Indian languages, and Language Detection recognises romanised Indic text, such as Hindi typed in Latin letters, which is how many customers write in chat. Build a support-ticket triage that detects the language, translates to English and returns the customer's issue as JSON.

Implement information extraction solutions (10 to 15%)

Ingesting and indexing documents, images, audio and video; semantic, hybrid and vector search; enrichment skills; OCR in the RAG ingestion flow; and Content Understanding analyzers that produce structured or markdown output for agents.

Practise: index a folder of scanned PDFs with OCR in Azure AI Search, compare keyword, vector and hybrid results for the same question, then connect the index to your agent as a tool. Microsoft's region table shows Central India supporting AI enrichment, semantic ranker and agentic retrieval for Azure AI Search, so you can keep this lab in India.

Where your model runs: a question Indian clients ask

Foundry offers three families of deployment type, described in Deployment types for Microsoft Foundry Models:

Type	Where prompts are processed
Global	Any Azure region where the model is deployed
Data Zone	Anywhere inside a Microsoft data zone: US, EU or Asia Pacific
Standard (regional)	Only in the region of your deployment

Data stored at rest stays in the Azure geography in every case. Microsoft recommends starting with Global Standard for most workloads, because new models arrive there first, at the lowest price and with the broadest region coverage. If a client policy says prompts must be processed inside India, the Asia Pacific data zone does not meet it; you need a regional deployment in an Indian region where that model is offered, and quota there can be tighter. The exam tests this trade-off under "choose appropriate deployment options", so know which type answers which requirement.

Prerequisites and an eight-week plan for working developers

Microsoft lists no prerequisite certification. The audience profile asks for experience developing apps in Python and familiarity with generative AI and Azure services. If Azure itself is new to you, AI-901 first gives you a gentler start; our AI-901 vs AI-900 guide explains it.

The plan below assumes weekday evenings in IST plus one longer weekend session.

Week	Focus	Build by Sunday
1	Foundry setup, deployments, keyless auth	Python client using `DefaultAzureCredential`
2	RAG with Azure AI Search	Grounded answers over your own PDFs
3	Agents, tools, function calling, memory	Agent with one custom function
4	Multi-agent flows, approvals, evaluation	Two-agent workflow with human approval
5	Vision, image and video generation	Alt-text generator and injection test
6	Text, speech and translation	Multilingual ticket triage
7	Information extraction	Form or invoice extractor with Content Understanding
8	Security, monitoring, practice	Tracing turned on, practice assessment, weak areas revised

Keep costs down by creating one resource group per week, deleting it on Sunday night and setting a budget alert on day one.

Exam day and renewal

Microsoft Learn is open during the exam. Role-based exams let you browse learn.microsoft.com, but the timer keeps running. Use it to check a parameter name, not to research every question.
Case studies are possible. If you get one, read the requirements first, then the questions, then return to the details you need.
Practice assessment: Microsoft's practice assessment for this certification is on AI Skills Navigator, and you must sign in to launch it.
Online proctoring: use a personal laptop if you can. Microsoft notes that work computers often have software that stops the OnVUE browser from launching.
Renewal: the certification expires after a year, and you renew it with a free online assessment. See our renewal guide for India.

How CertTulen Academy can help

CertTulen Academy, a Microsoft Training Services Partner, runs the AI-103 Azure AI Apps and Agents Developer course as live, MCT-led online training with hands-on Foundry labs that follow the five skill areas above. Professional certification courses include an exam voucher, mock practice tests and lab access; the course page has the details and INR pricing. 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.