AI-102 vs AI-103: The Azure AI Engineer Certification Was Replaced, Not Renewed

Exam AI-102 retired on 30 June 2026, and so did the credential it awarded — Microsoft Certified: Azure AI Engineer Associate. Its replacement, AI-103, awards a differently named credential: Azure AI App and Agent Developer Associate.

This is the detail people get wrong, because the AI-900 story trained everyone to expect the opposite. There is no transition path from the old certification to the new one: holding AI-102 does not shorten, discount or partially satisfy AI-103. If you want the current credential, you sit the full exam. Bookings could not be transferred either — an AI-102 booking had to be cancelled and rebooked as AI-103.

The rename is not cosmetic. "AI Engineer" described someone wiring up Azure AI services; "App and Agent Developer" describes someone building agentic applications on Microsoft Foundry. The exam content moved to match. Below is the new domain split, what it means in practice, and how it relates to the fundamentals exam beneath it. Gnoseed does not sell, proctor or guarantee these exams — always confirm the current details on Microsoft Learn before booking. For the fundamentals tier below this one, see what changed between AI-900 and AI-901.

AI-103 publishes five skills-measured domains: 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%) and Implement information extraction solutions (10–15%).

Read those weights together and the emphasis is clear: roughly two thirds of the paper is planning, managing and building generative and agentic solutions. Vision, text and extraction — the three areas that dominated AI-102 — now share about a third between them.

The planning domain is the one most likely to surprise an AI-102 holder. It covers choosing Foundry services and models for a task, designing infrastructure and deployment options, managing quotas, scaling, rate limits and cost, monitoring drift, safety events and grounding quality, and configuring managed identity, private networking and keyless credentials. That is an architecture and operations skill set, not a service-integration one.

Microsoft describes the AI-103 candidate as an Azure AI engineer who builds, manages and deploys agents and AI solutions that take advantage of Microsoft Foundry, with experience developing apps in Python and familiarity with general AI, generative AI and Azure services.

The listed responsibilities are worth reading as a checklist: planning and managing Azure AI solutions, implementing generative and agentic solutions, and implementing computer vision, text analysis and information extraction. Collaboration with solution architects, data scientists, DevOps and cloud security engineers is called out explicitly, which tells you the exam expects opinions about deployment, cost and security, not only about APIs.

Responsible AI is embedded rather than bolted on. Safety filters and guardrails, risk detection, evaluators and safety evaluations, trace logging and provenance metadata, and governing agent behaviour with oversight modes and tool-access controls all sit inside the planning domain.

There is no prerequisite — you can sit AI-103 without AI-901. But the two exams now share a spine, and that changes the calculus compared with the AI-900 and AI-102 era, where the fundamentals exam covered largely different ground.

AI-901 asks you to build a lightweight app or a single-agent solution in Foundry. AI-103 asks you to design the infrastructure, orchestrate multi-agent solutions, evaluate them, monitor drift and grounding, and secure them with managed identity and private networking. The vocabulary is the same; the depth is not.

If Foundry is new to you, the fundamentals exam is a cheap way to build the vocabulary before paying for the associate one. If you already ship on Foundry daily, go straight to AI-103 — you would be paying for a certification whose content you already work with.

More than the rename suggests. Computer vision, text analysis and information extraction are all still there, and if you prepared those for AI-102 the underlying concepts hold — OCR, layout analysis, entity extraction, speech to text and text to speech have not changed shape.

What changed is how you are expected to reach them. Work that used to mean calling a per-workload service now runs through multimodal models and Azure Content Understanding, and the products carry new names: Azure AI Services are Foundry Tools, and Azure AI Foundry is Microsoft Foundry.

The genuinely new material is the agentic half — prompt agents versus hosted agents, toolboxes, the Responses API, multi-agent orchestration, agent identity and agent observability. Budget most of your preparation there, because none of it appeared on AI-102.