PracticeVideosBlogFlashcardsEspañol
Cheat sheet

AI-103 Cheat Sheet

Practice QuestionsStudy Guide

Plan + Manage

25-30%of exam

FoundryModel ChoiceDeploymentsSecurityMonitoring

Generative + Agents

30-35%of exam

RAGFunction CallingAgent ToolsEvaluationTracing

Computer Vision

10-15%of exam

MultimodalCaptioningVQAImage GenerationVisual Safety

Text Analysis

10-15%of exam

EntitiesSentimentTranslationSpeechStructured JSON

Information Extraction

10-15%of exam

Azure AI SearchVector SearchHybrid SearchOCRContent Understanding

Quick Facts

Exam
AI-103
Credential
AI Apps Associate
Platform
Microsoft Foundry
Pass
700/1000
Time
120 min
Questions
40-60 typical
Price
$165 US
Blueprint
Apr 16 2026

Observability

Trace tokens latency safety

TracesTokensLatencySafety

Identity vs Key

Managed identity

  • Keyless
  • RBAC
  • Rotates automatically

API key

  • Secret value
  • Must protect
  • Manual rotation

Identity beats secrets

Solution Picker

  1. Need keyless auth→Managed identity(Security)
  2. Need private traffic→Private endpoint(Network)
  3. Need grounded answers→RAG(Sources)
  4. Need retrieval index→AI Search(Grounding)
  5. Need visual reasoning→Multimodal model(Vision)
  6. Need document fields→Document Intelligence(Extraction)
  7. Need unsafe blocking→Content filters(Safety)
  8. Need tool actions→Agent Service(Agents)

Foundry Planning

Foundry
AI app platform
Hub
Shared governance
Project
App workspace
Model catalog
Model selection
Deployment
Callable model endpoint
Quota
Capacity limit
TPM
Token throughput
PTU
Provisioned throughput

Filter vs Shield

Content filter

  • Classifies content
  • Blocks categories
  • Input/output

Prompt shield

  • Injection defense
  • Jailbreak detection
  • Indirect attacks

Content vs attack

Security + Ops

Managed identity
Keyless auth
Private endpoint
Private network path
Key Vault
Secret storage
RBAC
Role authorization
Content filter
Moderates content
Prompt shield
Injection defense
Monitor
Production signals
Trace
Step evidence
Safety event
Risk signal

RAG

Retrieve before generate

RetrieveGroundCiteAnswer

RAG vs Fine-tuning

RAG

  • Adds facts
  • Uses sources
  • Easier updates

Fine-tuning

  • Changes behavior
  • Domain style
  • Training data

Facts vs behavior

Generation Picker

  1. Need exact schema→Structured outputs(JSON)
  2. Need API action→Function calling(Tools)
  3. Need low latency→Streaming(UX)
  4. Need bulk runs→Batch API(Offline)
  5. Need domain tone→Fine-tuning(Behavior)
  6. Need current facts→Grounded RAG(Retrieval)
  7. Need quality score→Evaluators(Metrics)
  8. Need debug path→Tracing(Evidence)

Generative Patterns

RAG
Retrieve then generate
Grounding
Source-backed answer
Embeddings
Vector representation
Function calling
Tool invocation
Structured outputs
Schema-constrained JSON
Streaming
Token-by-token response
Prompt flow
Workflow orchestration
Fine-tuning
Model adaptation
Batch API
Offline bulk jobs
Evaluation
Quality scoring

Agent Core

Agents need role tools memory

RoleToolsMemoryGuardrails

Function vs Structured

Function

  • Calls tool
  • Takes action
  • Uses schema

Structured

  • Returns JSON
  • No tool call
  • Schema output

Act vs format

Agent Building

Agent role
Behavior definition
Goal
Task objective
Tool schema
Callable contract
Memory
Conversation state
Thread
Conversation container
File search
Knowledge tool
Code interpreter
Analysis tool
OpenAPI tool
External API
Multi-agent
Orchestrated agents
Approval flow
Human checkpoint

Agent vs Flow

Agent

  • Chooses tools
  • Uses memory
  • Goal-driven

Flow

  • Fixed steps
  • Deterministic path
  • Orchestrated workflow

Adaptive vs fixed

Quality Observability

Groundedness
Uses provided sources
Relevance
Answers request
Fabrication
Unsupported claim
Latency
Response time
Token analytics
Usage breakdown
Drift
Behavior change
Feedback
User quality signal
Canary
Controlled rollout

Vision Workloads

Multimodal
Text plus media
Captioning
Image description
VQA
Visual question answering
Alt text
Accessible description
Image generation
Text to image
Inpainting
Masked image edit
Video analysis
Segment understanding
Object detection
Locate items
Visual policy
Image safeguards

Text + Speech

Entities
Named facts
Topics
Theme extraction
Summaries
Condensed text
Sentiment
Tone detection
PII
Sensitive data
Translator
Language conversion
STT
Speech to text
TTS
Text to speech
SSML
Speech markup
Custom speech
Domain adaptation

Search Stack

Vector plus keyword wins

VectorBM25SemanticCitations

Vector vs Hybrid

Vector

  • Semantic similarity
  • Embeddings
  • Concept match

Hybrid

  • Vector + keyword
  • Better recall
  • BM25 included

Similarity vs combined

Extraction Picker

  1. Need similarity search→Vector search(Embeddings)
  2. Need keyword plus semantic→Hybrid search(Recall)
  3. Need result rerank→Semantic ranker(Relevance)
  4. Need image text→OCR(Read)
  5. Need form fields→Document Intelligence(Forms)
  6. Need multimodal parse→Content Understanding(Media)
  7. Need permissions→Security trimming(Access)
  8. Need source proof→Citations(Trust)

DocIntel vs Content

Doc Intelligence

  • Forms/docs
  • Layout fields
  • OCR pipeline

Content Understanding

  • Multimodal
  • Analyzers
  • Agent-ready output

Documents vs multimodal

Document Extraction

OCR
Text from images
Layout
Document structure
Fields
Extracted values
Tables
Grid extraction
Document Intelligence
Form extraction
Content Understanding
Multimodal extraction
Analyzer
Extraction configuration
Markdown output
Reasoning-friendly text
Grounded representation
Trusted downstream data

Common Traps

Agent vs chatbot

Agent uses tools ≠ Chatbot answers only

RAG vs memory

RAG fetches sources ≠ Memory tracks conversation

Filter vs shield

Filters classify content ≠ Shields block injection

Key vs identity

Keys are secrets ≠ Identity is keyless

Vector vs semantic

Vector finds similarity ≠ Semantic reranks text

Evaluation vs monitoring

Evaluation scores quality ≠ Monitoring watches production

Vision text risk

Image text can inject ≠ Treat media as input

Last Minute

  1. 1.Use April 2026 weights
  2. 2.Gen+agents largest 30-35%
  3. 3.Plan domain 25-30%
  4. 4.RAG needs search index
  5. 5.Agents need tool controls
  6. 6.Use managed identity
  7. 7.Trace tokens and latency
  8. 8.Evaluate groundedness and safety
  9. 9.Vision includes prompt injection
  10. 10.Extraction includes OCR

Microsoft Azure AI App and Agent Developer (AI-103)

Practice QuestionsStudy GuideCheat SheetFlashcards1 video1 blog

AI-103 Exam Prep and Practice 2026-2027: Azure AI Agent Developer Certification Study Guide with 680 Questions, Three Full Exams, and Complete Microsoft Foundry Coverage

$9.99 · Buy on Amazon

Take courses on Udemy
Same family resources

Explore More Microsoft Azure Certifications

Continue into nearby exams from the same family. Each card keeps practice questions, study guides, flashcards, videos, and articles in one place.

Azure AZ-104

Cheat SheetPractice QuestionsStudy GuideFlashcards4 videos5 blogs

Azure AZ-900

Cheat SheetPractice QuestionsStudy GuideFlashcards3 videos5 blogs

Azure AI-900

Cheat SheetPractice QuestionsStudy GuideFlashcards2 videos4 blogs

Azure DP-700

Cheat SheetPractice QuestionsStudy GuideFlashcards1 video4 blogs

Azure DP-600

Cheat SheetPractice QuestionsStudy GuideFlashcards1 blog

Azure DP-900

Cheat SheetPractice QuestionsStudy GuideFlashcards1 blog

More From This Family

Videos and articles for deeper review.

VideoMicrosoft AI-103 Exam Guide 2026: Azure AI Apps and AgentsThis examVideoAI-900 Exam Guide 2026: FREE Azure AI Fundamentals Study PlanVideoAZ-104 Study Plan 2026: Pass Azure Administrator ExamVideoAZ-500 Exam Guide 2026: Azure Security Before RetirementArticleMicrosoft AI-103 Exam Guide 2026: Azure AI Apps and Agents16 min readArticleFREE AZ-700 Exam Guide 2026: 5 Domains, $165, Pass Score 700/100017 min readArticleFREE DP-300 Exam Guide 2026: 5-Domain Skills Outline + Pass on First Try16 min readArticleAZ-104 Study Plan 2026: Pass Azure Administrator Exam17 min read