AI-103 is the exam behind Microsoft Certified: Azure AI Apps and Agents Developer Associate. It replaced the Azure AI Engineer Associate track when AI-102 retired on 30 June 2026, and it is a different job rather than a rewrite of the old one. AI-102 asked whether you could wire up Azure AI services. This asks whether you can build and run an agent. Five domains. Implement generative AI and agentic solutions is the largest at 30 to 35%, and it runs from deploying models through RAG, agent tools and function calling, multi-agent orchestration, approval flows and agent observability. Plan and manage an Azure AI solution takes 25 to 30% and covers model and service selection, deployment options, quotas and rate limits, managed identity and keyless access, and responsible AI including content filters and evaluators. The three implementation domains take 10 to 15% each: computer vision, text analysis and speech, and information extraction with Azure AI Search and Content Understanding. Microsoft publishes the time and not the item count: 120 minutes, 700 out of 1000 to pass, $165, proctored through Pearson VUE. These tests run 55 questions, which is where a paper of this shape lands in that window. Expect Python. The audience profile says so outright, and the questions that show you code show you the Foundry SDK, the OpenAI Responses API and workflow definitions in YAML. You are asked to complete them, not to write them from scratch. Formats matter here more than on most exams. Roughly a third of what the question site shows is dropdown grids and drag-and-drop matching rather than plain multiple choice, plus a case study whose scenario carries several questions. These tests are built in the same mix, because a pack that is all four-option multiple choice trains you for a paper you are not going to sit.