AI-102 to AI-103: What Changed in Microsoft's New Azure AI Exam (2026)

Microsoft retired Exam AI-102 on June 30, 2026, and replaced it with Exam AI-103: Developing AI Apps and Agents on Azure. Passing AI-103 earns the new Microsoft Certified: Azure AI Apps and Agents Developer Associate credential. The new exam keeps the familiar foundations — planning Azure AI solutions, retrieval-augmented generation (RAG), vision, language and document extraction — but moves the weight decisively toward generative AI and AI agents built with Microsoft Foundry, which now make up the largest domain at 30–35% of the exam.

This guide compares the two exams domain by domain, using Microsoft's own study guides, so you know exactly what carries over from AI-102 and what you need to learn fresh.

Key facts at a glance

Exam

AI-103: Developing AI Apps and Agents on Azure

Credential

Microsoft Certified: Azure AI Apps and Agents Developer Associate (intermediate)

Replaces

AI-102 (Microsoft Certified: Azure AI Engineer Associate), retired June 30, 2026, 11:59 PM CST

Skills measured

Version as of April 16, 2026 — 5 functional groups

Biggest domain

Implement generative AI and agentic solutions (30–35%)

Duration

120 minutes, proctored; may include interactive components

Passing score

700 or greater

Programming language

Python (AI-102's profile named Python and C#)

Languages

English, Chinese (Simplified), Chinese (Traditional), French, German, Japanese, Korean, Italian, Portuguese (Brazil), Spanish

Official practice

Free Practice Assessment on Microsoft AI Skills Navigator

Timeline: from AI-102 to AI-103

How the skill domains changed

AI-102 had six functional groups; AI-103 has five. Generative AI and agents — two separate, smaller domains on AI-102 — are merged into one domain that is now the heaviest on the exam. Planning and managing solutions also gained weight, while language and knowledge-mining topics were slimmed down.

AI-102 versus AI-103 domain weights

AI-102 domain (retired)	Weight	AI-103 domain	Weight	Change
Plan and manage an Azure AI solution	20–25%	Plan and manage an Azure AI solution	25–30%	▲ +5 pts
Implement generative AI solutions + Implement an agentic solution	15–20% + 5–10%	Implement generative AI and agentic solutions	30–35%	▲ merged, +5–10 pts
Implement computer vision solutions	10–15%	Implement computer vision solutions	10–15%	Same weight, new focus
Implement natural language processing solutions	15–20%	Implement text analysis solutions	10–15%	▼ −5 pts
Implement knowledge mining and information extraction solutions	15–20%	Implement information extraction solutions	10–15%	▼ −5 pts

What's new in AI-103

Comparing the two study guides line by line, these skills appear in AI-103 but were absent or only implied in AI-102:

• Agents as a core skill, not a 5–10% add-on. Defining agent roles, goals and tool schemas; agents that combine retrieval, function calling and conversation memory; orchestrated multi-agent solutions; autonomous or semiautonomous workflows with safeguards and approval flows; and monitoring and error analysis for deployed agents.
• Governance of agent behavior. Oversight modes, constraints and tool-access controls, plus auditing through trace logging, provenance metadata and approval workflows.
• Evaluation and observability. Evaluating apps for fabrications, relevance, quality and safety; tracing, token analytics, safety signals and latency breakdowns; chain-of-thought evaluations and self-critique loops.
• Operating AI at scale. Managing quotas, scaling, rate limits and cost footprints for model and agent workloads, and monitoring drift, grounding quality and search index health.
• Keyless, private security. Managed identity, private networking, keyless credentials and role policies (AI-102 still listed managing and protecting account keys).
• Generative vision. Generating images and videos from text prompts and reference media, inpainting and mask-based edits, video editing workflows, and multimodal question answering grounded in visual evidence.
• Multimodal safety. Detecting indirect prompt injection hidden in text embedded in images, and enforcing visual policy rules such as watermarks and brand usage.
• Speech for agents. Speech as an agent modality, multimodal reasoning from audio inputs and speech translation with language models.
• Azure Content Understanding runs through vision, document extraction and RAG ingestion — including single-task and pro-mode pipelines and analyzers that output structured or markdown results.

What's no longer listed

These AI-102 skills do not appear in the AI-103 skills outline. Related topics can still show up — Microsoft notes that the bullets illustrate rather than limit the exam — but they are no longer named objectives:

Already certified on AI-102?

If you earned the Azure AI Engineer Associate certification, it stays on the transcript in your Microsoft Learn profile. However, Microsoft has retired both the certification and its renewal assessment, so it can no longer be renewed. To hold a current Microsoft AI engineer credential, you need to pass AI-103 and earn the Azure AI Apps and Agents Developer Associate. Your AI-102 knowledge is a strong head start: RAG, semantic and vector search, content safety, OCR, translation and speech-to-text all carry over. Focus your study time on agents, evaluation, observability and generative vision.

AI-103 readiness checklist

Tick what you can already do — your progress is saved in this browser

0 of 15 ready

Plan and manage (25–30%)

Choose between LLMs, small language models and multimodal models for a task Configure managed identity and keyless access to Foundry models Manage quotas, rate limits and cost for model and agent workloads

Generative AI and agents (30–35%)

Build an agent with retrieval, function calling and conversation memory Orchestrate a multi-agent solution with approval flows Evaluate an app for fabrications, relevance and safety, and set up tracing

Computer vision (10–15%)

Generate and edit images (inpainting, mask-based edits) from prompts Analyze images and video with multimodal models and Content Understanding Mitigate prompt injection hidden in text inside images

Text analysis (10–15%)

Extract entities, summaries and structured JSON with generative prompting Add speech-to-text and text-to-speech to an agent Translate text with Azure Translator or LLM-powered flows

Information extraction (10–15%)

Configure hybrid, semantic and vector search for grounding Build a RAG ingestion flow with OCR and enrichment skills Extract fields and layout from documents with Content Understanding

Gaps found? Test yourself on exam-style AI-103 questions in our free 10-question preview — it includes a mini practice exam.

How to prepare for AI-103

1. Start with the official study guide and its skills outline — it is the definitive list of what can be tested.
2. Get hands-on in Microsoft Foundry. Build at least one RAG app and one agent with tools and memory; most of the new objectives are things you do, not definitions.
3. Weight your time like the exam does. Roughly a third on generative AI and agents, a quarter on planning, securing and monitoring, and the rest across vision, text analysis and information extraction.
4. Take Microsoft's free Practice Assessment on AI Skills Navigator to calibrate.
5. Drill exam-style questions and review why each wrong option is wrong — especially for scenario and case-study questions, which reward careful reading of requirements.

FAQ

Is AI-102 retired?

Yes. Microsoft retired Exam AI-102 on June 30, 2026, at 11:59 PM Central Standard Time. It can no longer be taken.

What replaced AI-102?

Exam AI-103: Developing AI Apps and Agents on Azure, which earns the Microsoft Certified: Azure AI Apps and Agents Developer Associate credential.

Is my Azure AI Engineer Associate certification still valid?

It stays on your Microsoft Learn transcript, but the certification and its renewal assessment are retired, so it cannot be renewed. To hold a current credential, take AI-103.

How different is AI-103 from AI-102?

The biggest change is that generative AI and agents are merged into one domain worth 30–35%, up from 20–30% across two domains. Planning and managing gains weight (25–30%), while language and knowledge-mining topics shrink to 10–15% each and computer vision shifts toward image and video generation and multimodal understanding.

Is AI-103 Python-only?

Microsoft's audience profile for AI-103 asks for experience developing apps with Python. AI-102's profile named both Python and C#.

How long is the AI-103 exam and what score do I need?

You have 120 minutes, and you need a score of 700 or greater to pass. The exam is proctored and may include interactive components.

Sources

• Microsoft Learn — Study guide for Exam AI-103 (skills measured as of April 16, 2026)
• Microsoft Learn — Study guide for Exam AI-102 (skills measured as of December 23, 2025; retirement notice)
• Microsoft Learn — Azure AI Apps and Agents Developer Associate (duration, languages, practice assessment)
• Microsoft Learn — Exam and assessment lab retirement (what happens to earned certifications)

Last checked against Microsoft Learn on October 11, 2026. Exam details change — always confirm on the official pages before you book.

Related guides

• Machine Learning Certifications Guide 2026
• AI Certifications for Forward Deployed Engineers
• Cloud Certification Roadmap 2026