Complete AI-103 Study Guide 2026: Azure AI Apps and Agents Developer

Quick answer: AI-103 is Microsoft's new 2026 associate-level AI certification, replacing AI-102 which retired June 30, 2026. It earns you the Azure AI Apps and Agents Developer Associate credential. The exam has 5 domains with a heavy focus on Azure AI Foundry, generative AI, and autonomous AI Agents. Passing score is 700/1000. SkillTech Club offers a full AI-103 course.

Who is AI-103 for?

AI-103 is an associate-level certification — it sits one level above the fundamentals (AI-901) and is aimed at developers and AI engineers who want to build AI-powered applications on Azure. You should consider AI-103 if you:

• Are a developer with Python or C# experience looking to move into AI
• Were studying for AI-102 — AI-103 is the direct replacement
• Want to build AI agents, RAG applications, or Azure OpenAI-powered apps
• Are an AI/ML engineer who wants an Azure-specific credential
• Want to work with Azure AI Foundry — Microsoft's platform for building AI applications

AI-103 exam at a glance

AI-103 exam domains — complete breakdown

AI-103 has 5 exam domains. Here is what each one tests and exactly how much it weighs in the exam.

Domain 1 — Plan and Manage Azure AI Solutions (25–30%)

This is the foundation domain. It covers how to architect and manage AI solutions on Azure:

• Selecting the right Azure AI services for a given business requirement
• Planning Azure AI Foundry projects — hubs, projects, connections, deployments
• Managing authentication, access control, and security for AI services
• Monitoring AI solutions — logging, alerts, diagnostics
• Cost management and responsible AI governance
• Content safety — Azure AI Content Safety service, content filtering policies

Study tip: Focus on when to use Azure AI Foundry vs standalone Azure AI services. This distinction comes up repeatedly in this domain.

Domain 2 — Implement Generative AI and Agents (30–35%) — Largest domain

This is the most important domain and the biggest change from AI-102. It is worth up to 35% of your exam:

• Azure OpenAI Service — deploying and using GPT-4, GPT-4o, DALL-E, Whisper models on Azure
• Prompt engineering — writing effective system prompts, few-shot examples, chain-of-thought prompting
• Azure AI Foundry prompt flow — building, testing, and evaluating LLM pipelines
• AI Agents — what they are, how they work, how to build them using the Azure AI Agent Service
• Multi-agent orchestration — coordinating multiple agents to complete complex tasks
• Grounding — connecting LLMs to your own data to improve accuracy and reduce hallucination
• Evaluation — measuring AI output quality, safety, and performance in Azure AI Foundry

Study tip: This domain requires hands-on practice. Create a free Azure account and spend time in Azure AI Foundry — deploy a model, run a prompt flow, and test the evaluation tools. You cannot fully understand this domain from reading alone.

Domain 3 — Implement Computer Vision Solutions (10–15%)

This is the most familiar domain for AI-102 holders — much of the content carries over:

• Azure AI Vision — image analysis, object detection, spatial analysis, OCR
• Azure AI Custom Vision — training custom classifiers and object detectors
• Azure AI Face — face detection, verification, identification
• Azure AI Document Intelligence — form recognition, layout analysis, custom models
• Video Indexer — extracting insights from video
• New: multimodal capabilities — GPT-4o's ability to understand both text and images

Domain 4 — Implement Text Analysis Solutions (10–15%)

NLP services on Azure — again familiar for AI-102 candidates:

• Azure AI Language — sentiment analysis, entity recognition, key phrase extraction, summarisation
• Azure AI Translator — real-time translation, custom translator
• Azure AI Speech — speech recognition, speech synthesis, speaker identification
• Conversational Language Understanding (CLU) — building intent and entity recognition models
• Knowledge base solutions — question answering using Azure AI Language
• New: LLM-enhanced NLP — using Azure OpenAI for advanced text tasks that go beyond traditional NLP services

Domain 5 — Implement Information Extraction (RAG) (10–15%)

This is a new standalone domain that did not exist as a separate section in AI-102:

• RAG (Retrieval Augmented Generation) — connecting an LLM to a search index so it answers questions based on your specific documents
• Azure AI Search — creating indexes, running semantic search, vector search
• Vector embeddings — what they are and why they are used in AI Search
• Building a RAG pipeline in Azure AI Foundry — connecting Azure OpenAI + Azure AI Search
• Hybrid search — combining keyword search and vector search for better results
• Evaluating RAG quality — groundedness, relevance, coherence metrics

Study tip: RAG is 10–15% of the exam but many candidates underestimate it. Build a simple RAG pipeline in Azure AI Foundry as part of your hands-on practice — it will make the concepts click immediately.

6-week AI-103 study plan

This plan assumes you are starting fresh (no AI-102 background). If you have AI-102 knowledge, compress Weeks 3–4 and spend more time on Domains 2 and 5.

Week 1 — Plan and Manage Azure AI Solutions (Domain 1)

• Azure AI services overview — what exists, what each does
• Azure AI Foundry — hubs, projects, deployments, connections
• Security, authentication, and access control for AI services
• Azure AI Content Safety — content filtering, harm categories
• Responsible AI principles in the context of production AI applications

Week 2 — Generative AI Foundations (Domain 2 Part 1)

• Azure OpenAI Service — deploying models, chat completions API, embeddings API
• Prompt engineering — system prompts, temperature, max tokens, few-shot
• Azure AI Foundry prompt flow — creating a basic flow, testing it
• Grounding and RAG concepts — why LLMs hallucinate and how grounding helps

Week 3 — AI Agents and Multi-Agent Systems (Domain 2 Part 2)

• What AI agents are — reasoning loops, tool use, memory
• Azure AI Agent Service — creating and deploying agents
• Tool calling — giving agents the ability to call APIs, search the web, run code
• Multi-agent orchestration — coordinator agents, worker agents
• Evaluating agent behaviour and handling failures

Week 4 — Computer Vision and NLP (Domains 3 and 4)

• Azure AI Vision, Custom Vision, Face, Document Intelligence
• Azure AI Language, Translator, Speech services
• Conversational Language Understanding (CLU)
• LLM-enhanced NLP — when to use Azure OpenAI vs traditional Azure AI Language

Week 5 — RAG and Information Extraction (Domain 5)

• Azure AI Search — indexing, semantic search, vector search
• Vector embeddings — creating and using them with Azure OpenAI
• Building a full RAG pipeline in Azure AI Foundry
• Hybrid search strategies
• Evaluating RAG pipeline quality

Week 6 — Full Review and Exam Readiness

• Days 1–3: Full practice tests across all 5 domains
• Day 4: Targeted revision on lowest-scoring domains
• Day 5: Light review only — no new material
• Book and sit the exam

Essential hands-on practice for AI-103

AI-103 is a developer exam — unlike AI-901, you are expected to have practical experience. Here are the hands-on exercises you must complete before exam day:

1. Deploy a model in Azure AI Foundry — use the model catalogue to deploy GPT-4o
2. Build a prompt flow — create a simple chat flow with a system prompt and test it
3. Create an Azure AI Search index — upload a few documents and run semantic search queries
4. Build a basic RAG pipeline — connect Azure OpenAI to your search index and ask it questions about your documents
5. Create an AI agent — use the Azure AI Agent Service to build an agent with at least one tool (e.g., Bing grounding or code interpreter)
6. Run an evaluation — use Azure AI Foundry's evaluation features to assess your prompt flow's quality

A free Azure account gives you enough credits to complete all of these exercises. Sign up at azure.microsoft.com/free.

Top 8 AI-103 exam tips

1. Domain 2 is 30–35% — treat it as the whole exam. If you get Domain 2 right, you pass with marks to spare.
2. Know when to use Azure AI Foundry vs Azure Machine Learning — AI Foundry is for AI application development and LLM workflows. Azure ML is for training custom models. This distinction appears in many scenario questions.
3. Understand RAG conceptually — why it exists, how it solves hallucination, how chunking and vector search work. You will not be asked to write code, but you will be asked to design RAG pipelines.
4. Know the difference between agents and chatbots — a chatbot responds to a prompt. An agent reasons, plans, and takes actions. This is a conceptual question that trips people up.
5. Content safety is tested — know what Azure AI Content Safety does, what harm categories it covers, and how to apply content filters in Azure AI Foundry.
6. Practice with case studies — AI-103 includes case study questions where you read a scenario and answer multiple questions about it. These require applying knowledge across multiple domains at once.
7. Know the Azure SDK — you will not write code in the exam, but questions may show code snippets and ask what they do. Understand the Azure OpenAI SDK and Azure AI Projects SDK at a high level.
8. Responsible AI at the application level — AI-103 tests responsible AI in a more applied way than AI-901. Know how to implement safety guardrails in real AI applications.

What comes after AI-103?

• Want to become a cloud architect? Take AZ-305 — Azure Solutions Architect Expert
• Want to go into DevOps? Take AZ-400 — Azure DevOps Engineer Expert
• Want to build low-code AI agents with Copilot Studio? Take the Copilot Studio Masterclass

Enrol in the AI-103 course at SkillTech Club

SkillTech Club's AI-103 course is taught by Maruti Makwana, a Microsoft Certified Trainer (MCT) with 18+ years of experience. The course covers every AI-103 domain in depth — including all-new content on Azure AI Foundry, AI Agents, RAG, and multi-agent orchestration.

Summary

• AI-103 has 5 domains — Domain 2 (Generative AI and Agents) is 30–35% and is the most important
• Coming from AI-102: domains 3 and 4 are familiar; domains 2 and 5 are entirely new
• Hands-on practice in Azure AI Foundry is essential — not optional
• Recommended study time: 6–10 weeks from AI-102 background, 10–14 weeks fresh start
• Build a RAG pipeline and create an AI agent as part of your preparation