Microsoft Certified: Azure AI Apps and Agents Developer Associate

What the exam actually covers

This is the exam for building AI applications on Azure rather than for describing them. The published domains are planning and managing an Azure AI solution, implementing generative AI and agentic solutions, computer vision, text analysis, and information extraction. The centre of gravity has shifted toward agents: multistep reasoning workflows and multi-agent orchestration are called out explicitly, which is the main thing separating this paper from the AI-102 exam it replaces.

Microsoft Foundry is the platform throughout, and the exam assumes you have used it. Expect to reason about which resource a described workload needs, how to authenticate to it, how to handle its responses in code, and where responsible AI controls belong in a working implementation.

Python is assumed. Microsoft names experience developing applications in Python and familiarity with general AI, generative AI and Azure services, so this is not a paper you pass from documentation alone. The sitting runs 120 minutes, is proctored, and may include interactive components alongside multiple choice, which is normal for a Microsoft role-based exam and less common elsewhere. Microsoft does not publish a question count for it, and we record that as unpublished rather than estimating.

Who it suits

Engineers already building on Azure, particularly anyone whose work has moved from calling individual AI services toward assembling agents. Microsoft maps it to both the AI Engineer and Developer job roles and marks it intermediate.

It suits holders of the retired Azure AI Engineer Associate most directly, since this is the credential their path now leads to. The overlap is real on the vision and language domains, and the new ground is the agentic and generative work.

How it compares

Against the AWS Machine Learning Engineer - Associate, the split is application development versus pipeline operation. AWS tests training, deploying and operating models on SageMaker; this tests building applications and agents on top of models someone else trained. They answer different questions about you.

Against the Google Cloud Professional Machine Learning Engineer, this is the less demanding paper and the more application-oriented one. Google's expects several years of modelling experience and weights MLOps heavily.

The closest comparison inside this directory is the Databricks Generative AI Engineer Associate, which covers similar gen AI engineering ground on a different platform, and the IBM watsonx Generative AI Engineer, which adds model governance weight. Which one signals correctly depends entirely on the cloud your employer runs.

What it costs to hold

Microsoft does not publish a price on the certification page, stating only that it depends on the country or region where the exam is proctored. Microsoft's own exam FAQ gives US$165 as the typical associate figure, so that is what we record, at medium confidence, and your local price may differ once taxes are applied.

The term is one year. Renewal is free through a short unproctored open-book assessment taken within a six-month window before expiry, so the running cost is time rather than money. That cadence is faster than the two and three-year terms on the AWS, Google and Databricks credentials, which is worth planning for if you hold several.

Before you book

Confirm the exam is in general availability in your language. Microsoft announced it in beta in April 2026 with general availability expected that June, and the page now lists ten languages.

Check the AI-103 study guide for recent changes before buying any course. This certification is new enough that third-party material may still describe the AI-102 syllabus, which omits most of the agentic content this paper is built around.