Microsoft Certified – Azure AI Apps and Agents Developer Associate (AI-103)

Learning plan

About 100 hours of focused study, split into 5 phases. Pick your weekly pace and the timeline adjusts.

Estimated finish: 10 weeks

What the exam covers

Source: official exam guide. Weights can change, so check before you start.

Before you start

• Comfortable writing Python apps (virtual environments, async calls, environment variables, REST calls)
• An Azure subscription with permission to create Foundry, Azure AI Search and Storage resources. Budget a few dollars for model tokens
• AI-901 level knowledge of AI workloads helps but is not required

1. Weeks 1–3 · 22 h

   Plan, secure and govern a Foundry solution

   • Choose between LLMs, small language models, multimodal models and Foundry Tools for a task, and justify the choice on cost, latency and quality
   • Design the Azure infrastructure for an agent app - Foundry resource and project, Azure AI Search, Storage, Key Vault, private endpoints
   • Configure keyless authentication with managed identity and RBAC instead of API keys, and restrict network access
   • Manage quotas, rate limits and deployment types (standard, global, provisioned) and wire a Foundry project into a CI/CD pipeline
   • Configure content filters, guardrails and risk detection, and set up trace logging and approval workflows for agent oversight

   Use: AI-103 study guide · AI-103T00 course and self-paced paths · Microsoft Foundry documentation

2. Weeks 3–6 · 34 h

   Generative apps, RAG and agents (biggest domain)

   • Build a Python app that connects to a Foundry project and calls a deployed model with the Foundry SDK
   • Implement RAG over your own documents with Azure AI Search as the grounding source
   • Build an agent in Foundry Agent Service with function calling, a knowledge tool and conversation memory, then add an MCP or OpenAPI tool
   • Orchestrate a multi-agent solution with a human approval step before a side-effecting tool runs
   • Evaluate apps and agents for groundedness, relevance and safety, and set up tracing with token and latency breakdowns

   Hands-on: Build a support agent in Foundry Agent Service that answers from a product manual indexed in Azure AI Search, calls a Python function to look up order status, and must ask for approval before a refund tool runs. Run the built-in groundedness and relevance evaluators on 20 test questions.

   Use: Foundry Agent Service overview · Observability and evaluation in Foundry · MicrosoftLearning AI agents labs (GitHub) · Microsoft AI Agents for Beginners (GitHub)

3. Weeks 6–8 · 16 h

   Retrieval pipelines and information extraction

   • Ingest documents, images, audio and video into an Azure AI Search index with built-in and custom skills, including OCR and layout enrichment
   • Configure and compare keyword, vector, hybrid and semantic ranking for the same queries
   • Use Content Understanding analyzers to turn PDFs and forms into structured fields or markdown that agents can consume
   • Expose a retrieval pipeline directly as an agent tool

   Hands-on: Index 30 PDFs with the Import and vectorize data wizard, then run the same five questions as keyword, vector and hybrid plus semantic queries and record which mode returns the right chunk first.

   Use: Azure AI Search documentation · Hybrid search overview · Azure Content Understanding

4. Weeks 8–9 · 14 h

   Vision, text and speech

   • Generate and edit images (inpainting, mask-based edits) and generate video from prompts and reference media
   • Produce captions, alt text and visually grounded answers with a multimodal model, and analyse video segments with Content Understanding
   • Detect indirect prompt injection hidden as text inside images and filter unsafe visual content
   • Extract entities, sentiment and structured JSON with prompts or Foundry Tools, and translate text with Azure Translator
   • Add speech-to-text, text-to-speech and speech translation as an agent modality

   Use: Azure Speech documentation · Microsoft Generative AI for Beginners (GitHub)

5. Weeks 9–10 · 14 h

   Practice assessment and gap review

   • Take the free AI-103 practice assessment and map every miss to one of the five skill areas
   • Rebuild from memory any lab you could not finish, especially agent tools and keyless auth
   • Practise finding answers fast on learn.microsoft.com, since role-based exams let you open Learn during the test
   • Take one timed MeasureUp practice test if you want a full-length rehearsal

   Use: AI Skills Navigator practice assessment · MeasureUp official practice tests · John Savill's Technical Training (YouTube)

You're ready to book when…

• You can build a grounded agent with a custom function tool from an empty folder in under two hours
• You can explain when to use vector, hybrid or semantic search and how to secure the app with managed identity
• You can name the evaluator you would use for fabrication, relevance and harmful content
• You score 80% or higher on the official practice assessment on two separate attempts

Exam tips

• Role-based exams let you open Microsoft Learn in a split screen, but the clock keeps running, so use it only for lookups you can't answer.
• The exam runs 120 minutes. Taking a break locks every question you have already seen, so finish reviewing before you pause.
• Certification renewal is a free online assessment each year rather than a retake.