Microsoft retired the AI-102 exam and the Azure AI Engineer Associate certification on June 30, 2026. Its replacement is AI-103: Developing AI Apps and Agents on Azure, which leads to a new credential — Microsoft Certified: Azure AI Apps and Agents Developer Associate. The exam runs 120 minutes, requires a 700/1000 score to pass, costs $165 USD at standard US pricing, and shifts the center of gravity hard toward generative AI and agents: they're now the largest domain at 30–35% of the exam.
If you were preparing for AI-102, most study material published before spring 2026 is now aimed at a dead exam. This guide covers what AI-103 actually tests (from the official study guide), what changed from AI-102, what the retirement means if you already hold the old credential, and how to prepare.
What Is the AI-103 Certification?
AI-103 is Microsoft's intermediate-level exam for engineers who build, deploy, and manage AI applications and agents on Azure using Microsoft Foundry (the platform formerly known as Azure AI Foundry). Passing it earns the Azure AI Apps and Agents Developer Associate certification.
The name change from "AI Engineer" to "AI Apps and Agents Developer" is the tell. Microsoft didn't refresh the old exam — it rebuilt the role profile around what AI engineers are actually shipping in 2026: RAG pipelines, tool-calling agents, multi-agent orchestration, and production guardrails. The audience profile in the study guide is explicit: you should have experience developing apps in Python and be familiar with generative AI capabilities and Azure services. AI-102's C#-or-Python flexibility is gone from the profile; Python is the assumed language.
AI-102 Is Retired: What It Means for You
AI-102 and the Azure AI Engineer Associate certification retired on June 30, 2026 — and Microsoft retired the renewal assessment with it. That second part matters more than the headline:
- If you hold Azure AI Engineer Associate: the credential stays on your transcript until its printed expiration date, but there is no renewal path. When it lapses, it's gone. If you want a current Azure AI credential after that, you take AI-103.
- If you were mid-study for AI-102: stop. Registrations are closed. The good news is the overlap is real — plan-and-manage, vision, language, and knowledge-mining skills carry over — but you should re-weight your prep toward agents and Foundry immediately.
- If you're starting fresh: there's no decision to make. AI-103 is the only path, and frankly the better one — it matches the job descriptions being posted for AI engineers right now.
I went through the same transition cycle with clients when AI-900 was replaced by AI-901 this spring: the people who moved early got certified against a near-empty question-bank ecosystem while the exam was still uncompetitive on the hiring signal side. The same window is open for AI-103 right now.
AI-103 Exam Details
| Detail | AI-103 |
|---|---|
| Full name | Developing AI Apps and Agents on Azure |
| Certification earned | Microsoft Certified: Azure AI Apps and Agents Developer Associate |
| Level | Intermediate (Associate) |
| Duration | 120 minutes |
| Passing score | 700 / 1000 |
| Cost | $165 USD (varies by country) |
| Delivery | Pearson VUE, proctored; may include interactive components |
| Languages | English (localizations follow ~8 weeks after updates) |
| Prerequisites | None (Python experience strongly assumed) |
| Renewal | Free online assessment on Microsoft Learn, annually |
Two practical notes from the fine print: the exam "may have interactive components," which in recent Microsoft exams has meant lab-style tasks rather than pure multiple choice — budget your 120 minutes accordingly. And the official Practice Assessment isn't live yet; Microsoft says practice assessments usually appear within 8 weeks of an exam leaving beta, so don't wait on it.
Skills Measured: The Five Domains
From the official study guide (skills measured as of April 16, 2026):
| Domain | Weight |
|---|---|
| Plan and manage an Azure AI solution | 25–30% |
| Implement generative AI and agentic solutions | 30–35% |
| Implement computer vision solutions | 10–15% |
| Implement text analysis solutions | 10–15% |
| Implement information extraction solutions | 10–15% |
The first two domains are 55–65% of your score. Prep time should follow the weights — I'd put roughly two-thirds of your hands-on hours into Foundry solution management and the generative/agentic domain, and treat the last three domains as consolidation work.
Plan and Manage an Azure AI Solution (25–30%)
This is the architecture-and-operations domain: choosing the right Foundry services and models (LLMs vs. small language models vs. multimodal), designing deployment options, wiring Foundry projects into CI/CD, and managing quotas, scaling, and cost. It also carries the responsible AI load — safety filters, guardrails, risk detection, trace logging, and governing agent behavior with tool-access controls.
Don't skim the security bullets: managed identity, private networking, keyless credentials, and role policies are all named explicitly. Microsoft has been pushing keyless auth across Azure services for two years, and exam content follows platform pressure.
Implement Generative AI and Agentic Solutions (30–35%)
The headline domain, and the one with the least overlap with old AI-102 material. Expect to be tested on:
- RAG applications — retrieval-augmented generation end to end, including evaluating outputs for fabrications, relevance, and safety
- Agent construction — defining agent roles, goals, and tool schemas; integrating retrieval, function calling, and conversation memory
- Multi-agent orchestration — coordinated agent solutions and autonomous workflows with approval-flow safeguards
- Operationalization — prompt engineering, self-critique loops, tracing, token analytics, and latency breakdowns
If you've only consumed a chat completion API through a wrapper library, this domain will expose the gap. Build at least one tool-calling agent with memory against a real knowledge store before you book the exam.
The Remaining Three Domains (10–15% each)
Vision, text analysis, and information extraction survive from AI-102 but in compressed, generative-flavored form. Vision now includes image and video generation (inpainting, mask-based edits, prompt-driven video workflows) alongside classic analysis. Text analysis folds in speech as an agent modality. Information extraction centers on building retrieval and grounding pipelines with Content Understanding — which conveniently doubles as prep for the RAG material in the big domain.
AI-102 vs AI-103: What Actually Changed
| AI-102 (retired) | AI-103 | |
|---|---|---|
| Credential | Azure AI Engineer Associate | Azure AI Apps and Agents Developer Associate |
| Generative AI + agents | ~15% of exam | 30–35% (largest domain) |
| Platform framing | Azure AI services (per-service) | Microsoft Foundry (unified platform) |
| Languages | Python or C# | Python assumed |
| Image/video generation | Not covered | Covered (vision domain) |
| Duration | 100 minutes | 120 minutes |
| Renewal | Retired — no renewal path | Free annual Learn assessment |
The pattern matches every 2026 Microsoft exam refresh, including AI-901 replacing AI-900: less "identify which service does X," more "build and operate the thing." AWS made the same move with its Generative AI Developer – Professional exam — cloud vendors are converging on testing agentic builders, not service catalogs.
How to Prepare for AI-103
Weeks 1–2 — Foundry fundamentals and solution planning. Work through the AI-103 learning paths on Microsoft Learn, then set up a Foundry project yourself: deploy a model, configure a managed identity, and connect an app with the Foundry SDK. The plan-and-manage domain rewards people who have actually clicked through (and scripted) the setup.
Weeks 3–4 — the agentic core. Build a RAG application with hybrid search grounding, then extend it into an agent with at least two tools and conversation memory. Add evaluation: fabrication detection, relevance scoring, and trace logging. This maps directly onto the 30–35% domain, and it's the part you can't fake with flashcards.
Week 5 — the three small domains. Run the vision, speech, and Content Understanding quickstarts. Focus on multimodal understanding and document extraction pipelines rather than memorizing API signatures — the exam tests scenario judgment, and the weights cap how much any one service detail can cost you.
Week 6 — consolidate and book. Re-read the study guide bullet by bullet and close gaps. If the official Practice Assessment has shipped by the time you read this, use it as your readiness gate; if not, the exam sandbox at least de-risks the question formats.
Certification study tells you what to build; the fastest way to make it stick is building it. CloudaQube's hands-on AI labs cover the same RAG, agent, and guardrail patterns AI-103 tests — and if you're weighing where this credential fits in a broader plan, the AI engineer career path guide maps certifications to actual hiring signals.
Is AI-103 Worth Taking in 2026?
For working engineers targeting AI roles on Azure: yes, and the timing argument is stronger than the credential argument. The exam went GA in June 2026, the SERP and question-bank ecosystem around it is still thin, and every AI-102 holder is being funneled toward it as renewals lapse — demand for "AI-103" content and prep is about to spike while supply is still catching up. Getting certified early means holding the credential while it still differentiates.
The honest caveat: if your work is AWS-first, this exam validates the right skills on the wrong platform — the agentic skills transfer, the Foundry specifics don't. In that case the AWS GenAI Developer Professional is the equivalent bet on the other side of the fence.
