Microsoft AI-103 Exam Guide: Domains, Weights and Format

What AI-103 is, and what it replaced

Exam AI-103, Developing AI Apps and Agents on Azure, leads to the Microsoft Certified: Azure AI Apps and Agents Developer Associate credential. It replaced AI-102, which retired on 30 June 2026 after years as Microsoft's main AI engineering exam. The rename is not cosmetic: the role profile was rebuilt around what AI engineers actually ship now — retrieval pipelines, tool-calling agents, multi-agent orchestration and production guardrails — and the generative and agentic domain is now the largest single block of marks.

Exam format at a glance

AI-103 is proctored through Pearson VUE and may include interactive components as well as multiple choice. Microsoft scores on a scaled 1000-point range with 700 required to pass, which is not the same as answering 70% correctly — the scale accounts for form difficulty.

Exam:        Developing AI Apps and Agents on Azure
Code:        AI-103
Credential:  Azure AI Apps and Agents Developer Associate
Price:       $165 USD (varies by proctoring region)
Duration:    120 minutes
Questions:   not published; roughly 40-60 in practice
Passing:     700 of 1000, scaled
Renewal:     annually, via a free online assessment
Assumes:     Python, and familiarity with Azure AI services

The five domains and their weights

Microsoft publishes weights as ranges rather than fixed percentages. Two domains carry the majority of the marks between them — generative AI and agents, and planning and managing the solution — which is where your study time belongs.

Implement generative AI and agentic solutions .... 30-35%
Plan and manage an Azure AI solution ............ 25-30%
Implement computer vision solutions ............. 10-15%
Implement text analysis solutions ............... 10-15%
Implement information extraction solutions ...... 10-15%

Who the exam is for

The audience profile is an Azure AI engineer who builds, manages and deploys agents and AI solutions on Microsoft Foundry. Python is assumed rather than optional — the study guide names it explicitly — as is familiarity with generative AI concepts and Azure services generally. It is an associate-level exam, so it does not demand the production operating experience that a professional-tier exam like NVIDIA's NCP-AAI does, but it does assume you have built something.

How AI-103 differs from a vendor-neutral agentic exam

Roughly a third of AI-103 sits in computer vision, text analysis and information extraction — workloads a purely agentic certification does not test at all. That is the main thing to understand before choosing between AI-103 and something like NCP-AAI: AI-103 certifies an Azure AI engineer, of which agent work is now the largest part, rather than certifying agent engineering as such. If you build on Azure, that breadth is the point. If you do not, most of it will not transfer.

How to prepare efficiently

Study in weight order. Lock down the generative and agentic domain first, then planning and management, and treat the three lighter domains as one block worth roughly a third of the exam between them. Because the weights are ranges, the safest reading is the top of each range for the heavy domains and the bottom for the light ones. The fastest way to find your gaps is a domain-weighted readiness test that mirrors the published blueprint, so you learn which domains you are actually weak in rather than getting a flat percentage.

Exam tip

Do not treat this as AI-102 with a new number. The generative and agentic domain is now the single heaviest block at 30-35%, and it is weighted toward building agents — tool schemas, memory, multi-agent orchestration, approval flows and evaluation — rather than calling a model and formatting the result.