AI-103 — Microsoft Azure AI Apps and Agents Developer Associate Study Plan
A practical AI-103 study plan for Microsoft Azure AI Apps and Agents Developer Associate candidates, with 7-day, 14-day, 30-day, and 60/90-day schedules.
How to use this Study Plan
This Study Plan is for candidates preparing for the Microsoft Azure AI Apps and Agents Developer Associate (AI-103) exam, official exam code AI-103. It is designed for developers and cloud practitioners who need to turn limited study time into a realistic preparation schedule.
Use the current Microsoft AI-103 skills outline as your source of truth, then use this plan to organize your practice around the work you are likely to perform as an Azure AI apps and agents developer:
- Build AI-enabled applications using Azure AI services.
- Work with models, prompts, responses, evaluations, and safety controls.
- Design retrieval-augmented generation patterns with Azure AI Search or similar grounding approaches.
- Build agent workflows that use tools, data, and orchestration.
- Apply identity, security, monitoring, and responsible AI practices.
- Troubleshoot app behavior, grounding quality, authentication issues, and production readiness concerns.
This is an independent study planning page and is not affiliated with Microsoft.
Which plan should you use?
Choose the shortest plan that still gives you enough time for practice, not just reading.
| Time until exam | Best plan | Use this if | Main risk | Primary goal |
|---|---|---|---|---|
| 7 days | Final review sprint | You have already studied most topics or are rescheduling is not possible | Too much new material too late | Identify weak areas, drill scenarios, and stabilize timing |
| 14 days | Focused plan | You know Azure basics and have some AI or developer experience | Shallow coverage of agents, RAG, and responsible AI | Cover each major workstream once and complete timed practice |
| 30 days | Balanced plan | You can study most days and need structured coverage | Spending too long on reading without practice | Build concept knowledge, hands-on familiarity, and exam rhythm |
| 60 days | Full preparation path | You are new to Azure AI apps or need deeper hands-on work | Forgetting early topics | Learn, build, drill, and review in spaced cycles |
| 90 days | Extended full path | You are balancing work, travel, or a low weekly study budget | Low momentum | Build steadily with weekly practice and repeat review |
Build your AI-103 topic map first
Before starting any schedule, create a one-page topic map. Do not try to memorize every product detail. Organize your study around repeatable developer decisions.
| Workstream | What to practice | Questions to ask yourself |
|---|---|---|
| Azure AI platform basics | Resource selection, endpoints, authentication, SDK/API patterns, deployment choices | Which service fits the requirement? How does the app authenticate? What must be configured before code works? |
| Model and prompt development | Prompt structure, system instructions, parameters, response handling, prompt testing | How do I make output reliable, constrained, and testable? |
| AI agents | Agent instructions, tools, function calling, workflow orchestration, state, handoffs, failure behavior | When should an app use an agent instead of a simple model call? What tools can the agent use safely? |
| Retrieval and grounding | Indexing, chunking, embeddings, vector and keyword retrieval, citations, freshness | How does the app retrieve the right context and avoid unsupported answers? |
| Azure AI Search patterns | Index design, analyzers at a high level, filters, scoring, semantic/vector retrieval concepts | What retrieval strategy matches the data and user query? |
| Responsible AI and safety | Content filtering, evaluation, groundedness checks, privacy, abuse prevention, human review | How are unsafe, sensitive, or low-confidence outputs handled? |
| Security and governance | Managed identities, role-based access, secrets, network exposure, least privilege | What should not be hard-coded? Who or what needs access? |
| Observability and troubleshooting | Logs, traces, latency, failed calls, token/usage awareness, search quality diagnostics | Where would I look when the app gives wrong, slow, or unsafe answers? |
Daily practice rhythm
Use the same rhythm almost every study day. Consistency matters more than long, unfocused sessions.
Standard 90-minute study block
| Time | Activity | Output |
|---|---|---|
| 0-10 min | Quick recall | Write what you remember from yesterday without notes |
| 10-35 min | Focused concept review | One small topic: agent tools, search indexing, content safety, identity, etc. |
| 35-60 min | Hands-on or scenario practice | Build, trace, compare, or reason through a realistic task |
| 60-80 min | Practice questions | Timed set of 10-20 questions or scenario prompts |
| 80-90 min | Missed-question log | Record the cause, fix, and retry date |
If you only have 30 minutes
| Time | Activity |
|---|---|
| 5 min | Review yesterday’s error log |
| 15 min | Drill one narrow topic |
| 10 min | Answer 5-8 timed questions and explain each answer |
If you have 2-3 hours
Use two study blocks with a break. Make the second block practice-heavy. Avoid turning the entire session into documentation reading.
Diagnostic-first practice
Start with a diagnostic set before you feel ready. The goal is not a score to celebrate; it is a map of what to study.
| Diagnostic result | What it means | Next action |
|---|---|---|
| Misses are mostly terminology | You need service and feature recognition | Build a glossary and compare similar services |
| Misses are mostly scenario decisions | You need architecture practice | Drill “which service/pattern should I choose?” questions |
| Misses are mostly code/API flow | You need developer workflow practice | Trace request/response, auth, SDK, and error handling examples |
| Misses are mostly RAG quality | You need retrieval design practice | Review chunking, indexing, filtering, ranking, grounding, and citations |
| Misses are mostly security | You need identity and access review | Drill managed identity, roles, secrets, private access, and least privilege |
| Misses are mostly safety/governance | You need responsible AI review | Practice content filtering, evaluations, monitoring, and escalation paths |
Missed-question review method
A missed question is useful only if you classify why you missed it.
Create a log with these columns:
| Column | What to write |
|---|---|
| Date | When you missed it |
| Topic | Agents, RAG, Azure AI Search, security, evaluation, monitoring, etc. |
| Scenario clue | The phrase in the question that should have guided you |
| Your wrong choice | What you selected or assumed |
| Correct idea | The principle that makes the right answer right |
| Error type | Knowledge gap, misread, overthinking, weak service selection, weak troubleshooting |
| Fix | Read, lab, flashcard, diagram, or repeat question set |
| Retry date | 2-4 days later |
Use this rule:
- If you miss a topic once, log it.
- If you miss it twice, schedule a focused drill.
- If you miss it three times, do a hands-on walkthrough or draw the architecture from memory.
Hands-on practice priorities for AI-103
You do not need to build a production system, but you should understand the moving parts well enough to reason through scenarios.
| Practice area | Minimum hands-on goal |
|---|---|
| Model call workflow | Understand app request, authentication, prompt/instructions, model response, error handling, and logging |
| Prompt and response control | Compare outputs from different instructions and constraints |
| RAG workflow | Trace document ingestion, chunking, embedding, indexing, retrieval, prompt grounding, and citation behavior |
| Agent workflow | Understand agent instructions, tools, function calls, state, and guardrails |
| Search troubleshooting | Diagnose poor results caused by bad data, bad chunks, weak filters, missing metadata, or poor query design |
| Security | Avoid hard-coded secrets; understand identities, roles, and access boundaries |
| Safety and evaluation | Test for unsafe output, ungrounded output, irrelevant retrieval, and unexpected tool use |
| Operations | Review logging, monitoring, latency, failed calls, and cost/governance tradeoffs |
A useful mental model for many AI-103 scenarios:
user request
-> authenticate and authorize
-> classify intent and apply safety checks
-> retrieve or select tools if needed
-> ground the prompt with trusted context
-> call the model or agent
-> validate, filter, and format the response
-> log telemetry and handle errors
7-day final review sprint
Use this if your exam is one week away. Do not attempt to learn every Azure AI topic from scratch. Your job is to find weak areas, reduce careless errors, and practice under time pressure.
| Day | Main focus | Study actions | Practice target |
|---|---|---|---|
| 1 | Diagnostic and triage | Take a timed diagnostic set. Build your error log. Rank weak areas by frequency and severity. | 40-60 mixed questions or one partial mock |
| 2 | Azure AI app foundations | Review resource setup, authentication, SDK/API flow, endpoints, permissions, and deployment choices. | Drill service-selection and auth scenarios |
| 3 | RAG and Azure AI Search | Review indexing, chunking, embeddings, vector/keyword retrieval, filters, grounding, and citations. | Drill retrieval quality and troubleshooting questions |
| 4 | Agents and tool use | Review agent instructions, tool/function calling, orchestration, state, and safe tool boundaries. | Drill agent design scenarios |
| 5 | Responsible AI, security, and operations | Review safety filters, evaluation, monitoring, secrets, managed identity, access control, and failure handling. | Drill mixed governance and troubleshooting questions |
| 6 | Timed mock | Take a full timed mock or the longest timed set you can support. Review every missed and guessed item. | Full timed practice |
| 7 | Light final review | Review notes, error log, service comparison tables, and exam-day timing. Stop heavy new learning. | Short confidence set only |
7-day rules
- Stop adding new material after Day 5 unless it fixes a repeated miss.
- Do not spend Day 7 on deep documentation reading.
- Redo missed questions from Days 1-6, but explain the reasoning before checking the answer.
- Prioritize scenario recognition over memorizing obscure details.
14-day focused plan
Use this if you have two weeks and can study 60-120 minutes most days.
| Day | Focus | Actions |
|---|---|---|
| 1 | Diagnostic | Take a baseline timed set. Build the topic map and error log. |
| 2 | Azure AI platform setup | Review Azure AI services, resource organization, endpoints, authentication, and SDK/API flow. |
| 3 | Model usage patterns | Practice prompt structure, instructions, parameters at a conceptual level, response formatting, and error handling. |
| 4 | Evaluation and prompt improvement | Compare good and poor prompts. Review groundedness, relevance, quality checks, and iteration. |
| 5 | RAG fundamentals | Study ingestion, chunking, embeddings, indexes, retrieval, grounding, and citations. |
| 6 | Azure AI Search scenarios | Drill index design, metadata, filters, hybrid/vector retrieval concepts, and search troubleshooting. |
| 7 | Mixed review checkpoint | Take a timed mixed set. Update weak-area ranking. |
| 8 | Agent fundamentals | Review agent purpose, instructions, tool use, function calling, and workflow orchestration. |
| 9 | Agent safety and reliability | Practice tool boundaries, fallbacks, state, escalation, and preventing unsupported actions. |
| 10 | Security and identity | Review managed identities, role-based access, secrets, least privilege, and secure app patterns. |
| 11 | Responsible AI and governance | Review content safety, privacy considerations, monitoring, evaluation, and human review patterns. |
| 12 | Timed mock | Take a full timed mock or extended timed set. Review all misses and guesses. |
| 13 | Weak-area sprint | Drill the top three weak topics only. Redraw key workflows from memory. |
| 14 | Final review | Review error log, scenario clues, and timing strategy. Do a short warm-up set, not a heavy mock. |
14-day rules
- By Day 11, stop broad new-topic study.
- Day 12 is for timed performance, not passive review.
- Day 13 should be narrow. Fix repeated mistakes rather than adding new resources.
- Day 14 should feel calm and repetitive.
30-day balanced plan
Use this if you want a realistic mix of learning, hands-on review, and timed practice.
Weekly structure
| Week | Goal | Main outputs |
|---|---|---|
| Week 1 | Build foundation | Topic map, baseline diagnostic, service comparison notes, basic app flow understanding |
| Week 2 | RAG and search depth | Retrieval workflow diagram, Azure AI Search scenario drills, troubleshooting notes |
| Week 3 | Agents, safety, and security | Agent workflow diagram, tool-use decision table, responsible AI checklist |
| Week 4 | Exam readiness | Timed mocks, weak-area sprint, final review sheet |
30-day schedule
| Days | Focus | Actions |
|---|---|---|
| 1 | Baseline | Take diagnostic practice. Create error log and study calendar. |
| 2-3 | Azure AI app architecture | Review service selection, resource setup, endpoints, SDK/API patterns, and app integration. |
| 4-5 | Authentication and access | Drill managed identity, role-based access, secrets, secure configuration, and least privilege scenarios. |
| 6 | Mixed practice | Timed set on platform, auth, and app foundations. |
| 7 | Review day | Redo missed questions. Summarize common scenario clues. |
| 8-9 | Prompt and model usage | Practice instructions, output constraints, response handling, and prompt improvement. |
| 10-11 | RAG fundamentals | Review chunking, embeddings, indexing, retrieval, grounding, and citations. |
| 12-13 | Azure AI Search | Drill vector, keyword, hybrid retrieval concepts, filters, metadata, and ranking/troubleshooting scenarios. |
| 14 | Timed checkpoint | Take a timed mixed set. Update weak-area list. |
| 15-16 | Agents | Study agent instructions, tool/function calling, orchestration, state, and handoffs. |
| 17 | Agent troubleshooting | Practice failures: wrong tool, missing permission, bad instructions, unsafe action, poor grounding. |
| 18-19 | Responsible AI | Review content filtering, privacy, evaluation, groundedness, human review, and abuse prevention. |
| 20 | Monitoring and operations | Review logging, telemetry, latency, errors, reliability, and cost/governance tradeoffs. |
| 21 | Timed mock 1 | Take a full timed mock or extended timed set. Review deeply. |
| 22-23 | Weak area 1 | Drill your weakest topic with hands-on or diagram-based practice. |
| 24-25 | Weak area 2 | Repeat for the second weakest topic. |
| 26 | Weak area 3 | Repeat for the third weakest topic. |
| 27 | Timed mock 2 | Take another timed mock. Compare error patterns to Day 21. |
| 28 | Final content review | Review service comparisons, RAG flow, agent flow, security, and responsible AI. |
| 29 | Error-log cleanup | Redo missed and guessed questions. Explain answers out loud. |
| 30 | Light review | Short warm-up only. Prepare exam logistics and rest. |
30-day rules
- Use Weeks 1-3 to learn and build.
- Use Week 4 to perform and refine.
- Stop broad new material around Day 26.
- If your mock results are inconsistent, spend less time reading and more time reviewing missed-question patterns.
60/90-day full preparation path
Use this if you are starting earlier or need deeper Azure AI experience.
60-day path
| Phase | Days | Goal | What to complete |
|---|---|---|---|
| Phase 1 | 1-7 | Orientation and diagnostic | Official skills outline review, baseline diagnostic, topic map, study environment |
| Phase 2 | 8-17 | Azure AI app foundations | Resource patterns, identity, SDK/API flow, service selection, secure configuration |
| Phase 3 | 18-29 | Model, prompt, and evaluation practice | Prompt design, response handling, evaluations, safety checks, quality improvement |
| Phase 4 | 30-40 | RAG and Azure AI Search | Ingestion, chunking, embeddings, indexes, retrieval, grounding, citations, troubleshooting |
| Phase 5 | 41-49 | Agents and orchestration | Agent instructions, tools, function calling, workflow design, state, error handling |
| Phase 6 | 50-55 | Security, responsible AI, and operations | Least privilege, content safety, monitoring, logging, reliability, governance |
| Phase 7 | 56-60 | Final readiness | Timed mocks, error-log review, weak-area sprint, light final review |
90-day path
For 90 days, use the same phases but slow the pace and add repetition.
| Phase | Days | Add this for the 90-day version |
|---|---|---|
| Orientation | 1-10 | Build a stronger Azure AI glossary and service comparison sheet |
| Foundations | 11-25 | Add more hands-on setup and authentication troubleshooting |
| Model and prompt work | 26-40 | Create multiple prompt versions and compare quality, safety, and reliability |
| RAG and search | 41-60 | Practice more data shapes: FAQs, long documents, structured metadata, and stale content |
| Agents | 61-73 | Practice more tool-use scenarios and failure paths |
| Security and operations | 74-82 | Review monitoring, access boundaries, and production-readiness tradeoffs |
| Final readiness | 83-90 | Take timed mocks, review missed questions, and stop adding new material in the final week |
Weekly routine for 60/90 days
| Day type | Activity |
|---|---|
| 3 days per week | Learn or review one AI-103 workstream |
| 1-2 days per week | Hands-on or architecture scenario practice |
| 1 day per week | Timed question set |
| 1 day per week | Missed-question review and spaced repetition |
Timed mock exam strategy
Timed mocks are most useful after you have enough coverage to learn from the results.
| Plan | First timed mock | Second timed mock | Final timed practice |
|---|---|---|---|
| 7 days | Day 1 diagnostic or Day 6 mock | Day 6 if Day 1 was partial | Day 7 short warm-up only |
| 14 days | Day 7 checkpoint | Day 12 full mock | Day 14 short warm-up only |
| 30 days | Day 21 | Day 27 | Day 29 short targeted set |
| 60 days | Around Day 45-50 | Around Day 56 | Final 2-3 days light review |
| 90 days | Around Day 60-65 | Around Day 80-84 | Final week targeted review |
How to review a timed mock
Do not only count correct answers. For each missed or guessed question, mark one of these:
| Mark | Meaning | Fix |
|---|---|---|
| K | Knowledge gap | Review the topic and make a flashcard or note |
| S | Scenario misread | Underline the requirement, constraint, and clue words |
| D | Decision error | Compare two similar services or patterns |
| T | Troubleshooting gap | Trace the failure path step by step |
| R | Recall issue | Add to spaced review |
| C | Careless error | Slow down on qualifiers such as most, least, first, best, or not |
High-value AI-103 review checklists
Azure AI apps and model workflow
You should be able to explain:
- How an application authenticates to Azure services.
- Where prompts, instructions, and configuration belong.
- How the app handles model responses and errors.
- How to constrain output format when the consuming app needs structure.
- How to log enough information for troubleshooting without exposing sensitive data.
- How to choose between a simple model call, a RAG pattern, and an agent workflow.
RAG and Azure AI Search
You should be able to reason through:
- Why grounding is needed.
- How documents become searchable chunks.
- How embeddings and vector retrieval fit into the workflow.
- When metadata filters improve retrieval.
- Why citations or source references matter.
- How poor chunking, stale content, or missing metadata can reduce answer quality.
- How to troubleshoot irrelevant, incomplete, or unsupported answers.
Agents and tool use
You should be able to explain:
- What the agent is allowed to do.
- Which tools the agent can call.
- How tool inputs and outputs are validated.
- How state or conversation history affects behavior.
- How to prevent unsafe or excessive tool use.
- What fallback path is used when the agent cannot complete a task.
- When a deterministic workflow is better than an autonomous agent pattern.
Security, responsible AI, and operations
You should be ready for scenarios involving:
- Managed identities and least privilege.
- Avoiding hard-coded secrets.
- Role-based access decisions.
- Network exposure and secure configuration at a conceptual level.
- Content safety and filtering.
- Groundedness, relevance, and quality evaluation.
- Monitoring, logging, latency, failed calls, and reliability.
- Cost and governance tradeoffs without relying on memorized prices or quotas.
When to stop adding new material
Stopping new material is part of the plan. Late-stage studying should improve accuracy, not create confusion.
| Plan | Stop broad new material | What to do instead |
|---|---|---|
| 7 days | Day 5 | Review error log and drill weak topics |
| 14 days | Day 11 | Timed mock, weak-area sprint, final notes |
| 30 days | Day 26 | Mocks, missed questions, scenario review |
| 60 days | Final 7-10 days | Timed practice and memory consolidation |
| 90 days | Final 10 days | Timed practice, weak-area review, light hands-on recall |
Final-week rules
Follow these rules regardless of which schedule you used:
- Review the current Microsoft AI-103 skills outline one last time.
- Redo your most important missed questions.
- Practice explaining why wrong answers are wrong.
- Review diagrams for RAG, agent workflow, authentication, and monitoring.
- Do not start a large new lab in the final 48 hours.
- Do not rely on memorized product limits, prices, quotas, or dates.
- Sleep and pacing matter; avoid turning the final night into a cram session.
Exam-readiness checks
You are likely ready when you can do most of the following without notes:
| Readiness check | Yes/No |
|---|---|
| I can explain the difference between a direct model call, a RAG app, and an agent workflow. | |
| I can choose an Azure AI service or pattern from a scenario requirement. | |
| I can trace how data is indexed, retrieved, grounded, and returned to a user. | |
| I can identify common causes of poor retrieval or unsupported answers. | |
| I can reason through agent tool use, permissions, and failure handling. | |
| I can apply least privilege and avoid hard-coded secrets in app scenarios. | |
| I can recognize responsible AI controls and evaluation needs. | |
| I can complete timed mixed practice without rushing at the end. | |
| I can explain my missed-question patterns and how I fixed them. |
Practical next step
Start with a timed diagnostic set, even if you do not feel ready. Build your missed-question log, choose the 7-day, 14-day, 30-day, or 60/90-day path, and schedule your first timed mock now. Then use each practice session to improve one specific AI-103 decision skill: service selection, RAG design, agent workflow, security, safety, or troubleshooting.