Exam Tips & Strategy
The AI-102 is a proctored exam that tests your ability to write code integrating Azure AI services. Knowing the exam mechanics is as important as knowing the SDKs.
Exam Format
| Detail | Value |
|---|---|
| Number of questions | ~40–60 questions |
| Duration | 100–120 minutes |
| Passing score | 700 out of 1000 |
| Cost | $165 USD |
| Question types | Multiple choice, multiple answer, drag-and-drop, hot area, case study |
| Penalty for wrong answers | None — always answer every question |
| Can you go back? | Yes, within a section. No, between sections. |
This exam retires on June 30, 2026. Plan your preparation timeline accordingly. Microsoft may replace it with a new certification aligned to Azure AI Foundry and agentic AI patterns.
Study Strategy
Weeks 1–2: Plan & Manage + Generative AI (Challenges 01–20)
These domains cover 35–45% of the exam. Generative AI has been increased to 15–20% weight — this is the fastest-growing section. Focus on Azure OpenAI, RAG patterns, and prompt engineering.
Weeks 3–4: Computer Vision + NLP (Challenges 24–39)
These are heavily SDK-focused. Know the exact client initialization patterns, API calls, and response parsing for Vision, Language, and Speech services.
Week 5: Agents + Knowledge Mining (Challenges 21–23, 40–48)
AI Search skillsets, custom skills, and Document Intelligence are detail-heavy. Agent architecture is new (5–10%) but straightforward.
Week 6: Review + Practice
- Take the Free Practice Assessment
- Review the Coverage Matrix — any gaps?
- Redo Break & Fix scenarios from each challenge
Key Focus Areas
Generative AI (Increasing Weight)
The exam has increased Generative AI coverage. Know these deeply:
- RAG architecture: embeddings → vector search → prompt construction → completion
- Azure OpenAI deployment models: GPT-4o, GPT-4o-mini, text-embedding-ada-002
- Content filtering: configuring input/output filters, severity levels
- System/user/assistant message roles and when to use each
- Token management: counting tokens, managing context windows, chunking strategies
SDK vs REST Patterns
The exam tests both approaches. Know when each is appropriate:
- SDK: Production applications, type safety, retry logic built-in
- REST: Quick prototyping, language-agnostic, understanding what SDKs do underneath
Authentication Patterns
# Key-based (simple, for development)
from azure.core.credentials import AzureKeyCredential
client = SomeClient(endpoint, AzureKeyCredential(key))
# Identity-based (production, recommended)
from azure.identity import DefaultAzureCredential
client = SomeClient(endpoint, DefaultAzureCredential())
Common Name Change Traps
| Old Name | New Name | Notes |
|---|---|---|
| Cognitive Services | Azure AI services | Umbrella name changed |
| Form Recognizer | Document Intelligence | Complete rebrand |
| LUIS | CLU (Conversational Language Understanding) | Migrated to Azure AI Language |
| QnA Maker | Custom Question Answering | Now part of Azure AI Language |
| Anomaly Detector | Deprecated (Azure AI Metrics Advisor) | May still appear in older questions |
| Azure Cognitive Search | Azure AI Search | Name + features updated |
| Azure OpenAI Studio | Azure AI Foundry | Portal renamed |
If you see an old name in a question, mentally map it to the new service.
Azure OpenAI Knowledge
Key concepts the exam tests:
- Deployment vs Model: You deploy a model to get an endpoint. Same model can have multiple deployments.
- Token limits: GPT-4o supports 128K input tokens. Know approximate limits for each model family.
- Content filtering: Default filters are ON. Know how to customize severity thresholds.
- Responsible AI: When to use content filtering, groundedness detection, abuse monitoring
- Function calling: How to define tools and handle tool_calls in the response
Time Management
| Section | Suggested Time |
|---|---|
| First pass through all questions | 60–70 minutes |
| Review flagged questions | 20–25 minutes |
| Case studies (if present) | 20–25 minutes |
| Buffer | 5–10 minutes |
Tip: Don't spend more than 2 minutes on any single question in your first pass. Flag it and move on.
Day-of Checklist
- Photo ID ready (passport or government-issued)
- Test your computer and internet connection (for online proctoring)
- Clear your desk — nothing allowed except monitor, keyboard, mouse
- Close all applications except the exam browser
- Have water nearby (allowed for online exams)
- Review the Cheat Sheet one final time (not during the exam!)
- Arrive 15 minutes early for check-in
Useful Links
| Resource | Link |
|---|---|
| Try the exam interface | Exam Sandbox |
| Free practice questions | Practice Assessment |
| Schedule the exam | Pearson VUE |
| Official study guide | AI-102 Study Guide |
| Exam Replay offer | Exam Deals |
After You Pass 🎉
- Your certification appears on your Microsoft Learn profile within 24 hours
- You get a digital badge via Credly to share on LinkedIn
- The certification is valid for 1 year — renew for free by passing an online assessment
- Consider your next step: AI-050 (Azure AI Foundry), AZ-305 (Architect), or specialize with data engineering
Ready to start studying? Head to Challenge 01.