AI-103 Exam Overview

Who the exam is for

Knowing who an exam is written for tells you how its questions think. Exam AI-103, Developing AI Apps and Agents on Azure, leads to the Microsoft Certified: Azure AI Apps and Agents Developer Associate certification, at the intermediate level.

The candidate is an Azure AI engineer who builds, manages, and deploys agents and AI solutions that take advantage of Microsoft Foundry. You should have experience developing apps by using Python, and be familiar with the capabilities of general AI, generative AI, and Azure services.

That is why questions in this course are framed as decisions a builder makes: which service, which agent type, which deployment, and what goes wrong if you pick the neighbor.

The role has five responsibilities: planning and managing Azure AI solutions, implementing generative AI and agentic solutions, implementing computer vision solutions, implementing text analysis solutions, and implementing information extraction solutions.

The engineer does not work alone. In this role you collaborate with business stakeholders, solution architects, data scientists, DevOps engineers, and cloud security engineers to design, implement, and maintain AI solutions.

Keep the difference clear: those five groups are the people the candidate works with. The candidate is the Azure AI engineer who builds and deploys the solutions.

The five domains

The skills measured as of April 16, 2026 group into five domains, each with a weight range.

Read aloud, the same table:

• Plan and manage an Azure AI solution weighs 25 to 30 percent.
• Implement generative AI and agentic solutions weighs 30 to 35 percent, the largest share.
• Implement computer vision solutions weighs 10 to 15 percent.
• Implement text analysis solutions weighs 10 to 15 percent.
• Implement information extraction solutions weighs 10 to 15 percent.

The two that are easiest to swap are the top pair. Planning sits at 25 to 30 percent and generative AI one step above at 30 to 35 percent: five points apart, with generative AI ahead.

Reading the weights

The weights become a study plan once you add them up. Together, planning and generative AI account for 55 to 65 percent of the exam, while the three remaining domains share the same 10 to 15 percent band each.

Planning plus generative AI is the largest pair at 55 to 65 percent. Swapping either of them for computer vision, text analysis, or information extraction drops the pair to 50 percent at most, and two light domains together reach only 20 to 30 percent.

That does not make the light domains optional. At 10 to 15 percent each, computer vision, text analysis, and information extraction add up to 30 to 45 percent of the exam between them, more than either heavy domain alone.

A practical rule: give planning and generative AI the most hours, and still cover each light domain properly, because together they are a large share of the score.

What the questions cover

Each skill in the study guide comes with bullets. The bullets under each skill illustrate how it is assessed, and related topics may also be covered. Treat them as a map of the territory, not as a closed list.

Preview features deserve a note, because Foundry moves fast and many capabilities start in preview. Most questions cover features that are generally available (GA), but the exam may contain questions on preview features if those features are commonly used.

So a widely used preview capability is fair game, while an obscure one is unlikely to be the focus. Throughout this course, a feature in preview is marked as such so you know where it stands.

Time, score, and format

The numbers to know before test day:

• You have 120 minutes to complete the exam.
• A score of 700 or greater is required to pass.
• The exam is proctored, and it may include interactive components.

An exam sandbox lets you explore the exam environment and the question types before test day. Using it once removes surprises about the interface, so your 120 minutes go to the questions.

If a first attempt does not go well, there is a defined path back. If you fail, you can retake the exam 24 hours after the first attempt, and the waiting time for later retakes varies.

After you pass, the credential needs a yearly check-in. Microsoft associate certifications expire annually, and you can renew by passing a free online assessment on Microsoft Learn.

Languages and updates

The exam is offered in English, Chinese (Simplified), Chinese (Traditional), French, German, Japanese, Korean, Italian, Portuguese (Brazil), and Spanish.

Two rules matter when your language is not English. First, timing of updates: the English version is always updated first, and localized versions are updated approximately eight weeks later. Second, missing languages: if the exam is not available in your preferred language, you can request an additional 30 minutes to complete it.

Put the update rule into a scenario. The English version receives new skills, and a week later a candidate books the Japanese version for two weeks out. Three weeks after the English change, the Japanese exam is probably still on the previous skills list, because localized versions follow about eight weeks behind.

The extra 30 minutes is not part of that story. It is something you request when the exam is not offered in your preferred language at all, and Japanese is one of the offered languages.