AI-103: AI apps and agents on Azure

Objectives and progression

Six modules with fictional APS and project cases, explained practice, and an internal assessment of 24 decisions in 60 minutes. Includes architecture, quality, and operations decisions with documented version boundaries. AI-103 syllabus effective 2026-04-16. Current name: Azure AI Apps and Agents Developer Associate. Official exam of 120 minutes and score 700/1000, not equivalent to 70%, with no fixed question count inferred. Official Portuguese is Brazil. Syllabus pro-mode terminology is distinguished from the retired API and preview agentic alternative.

Audience: AI and application engineers, APS teams, and technical AI project managers.

Prerequisites: Azure and AI fundamentals, Python development, APIs, identity, and version control.

300 estimated study minutes

• Select services and deliver an observable solution with acceptance criteria.
• Combine context, structure, and evidence to produce usable answers.
• Orchestrate work with identity, state, and accountability boundaries.
• Generate, edit, and interpret media with explicit quality and lifecycle criteria.
• Preserve meaning, terminology, and certainty when transforming language.
• Build useful, traceable representations for agents and reconciliation.

Modules

1. Architecture, project, and release
2. Generative applications and evaluation
3. Agents, memory, and approvals
4. Visual and multimodal content
5. Text, speech, and operational communication
6. Ingestion, retrieval, and extraction

Continue learning

Data / AI

• AI-900: Azure AI Fundamentals (historical)
• AI-102: Azure AI Engineer Associate, historical
• DP-900: Azure Data Fundamentals

References and version

AI-103 skills measured2026-04-16; Microsoft Foundry documentation2026-09-30

• AI-103 objectives effective April16,2026 · 2026-09-30
• Azure AI Apps and Agents Developer Associate exam format · 2026-09-30
• Foundry SDK and endpoint contracts · 2026-09-30
• Deployment processing boundaries and capacity types · 2026-09-30
• Foundry workflows preview orchestration patterns · 2026-09-30
• Agent Framework approval request and response · 2026-09-30
• Memory scope identity and updates · 2026-09-30
• Memory preview lifecycle and limitations · 2026-09-30
• Schema constrained outputs and API differences · 2026-09-30
• Current image generation editing and DALL-E3 retirement · 2026-09-30
• Sora2 preview asynchronous video and remix · 2026-09-30
• Agentic preview migration from retired pro mode · 2026-09-30
• Analyzer schema workflow output and evidence configuration · 2026-09-30
• Adversarial safety evaluation · 2026-09-30
• Structure-aware document chunking · 2026-09-30
• Vector storage and compression tradeoffs · 2026-09-30
• Speech adaptation and evaluation · 2026-09-30
• Translator text and document workflows · 2026-09-30
• Summarization extractive and abstractive workflows · 2026-09-30
• Agent applications versions and deployment · 2026-09-30
• Foundry role scope and permissions · 2026-09-30
• Technical exam scaled scoring · 2026-09-30
• Foundry Tools authentication · 2026-09-30
• Private network and DNS access · 2026-09-30
• Classic Azure OpenAI quota rate limits and retries · 2026-09-30
• Generative solution evaluation and observability · 2026-09-30
• Agent function request application execution and output · 2026-09-30
• Retry limits idempotency and uncertain outcomes · 2026-09-30
• PII detection and redacted output · 2026-09-30
• SSML speech synthesis controls · 2026-09-30
• Speech recognition error evaluation · 2026-09-30
• Indexer source enrichment and output mappings · 2026-09-30
• Hybrid keyword and vector retrieval · 2026-09-30
• Semantic reranking of retrieved candidates · 2026-09-30
• Search security filtering by user access · 2026-09-30
• Retrieval augmented generation · 2026-09-30
• Direct and indirect prompt attacks · 2026-09-30
• Content Understanding multimodal structured extraction · 2026-09-30
• WAI alternative text decision tree · 2026-09-30