Build and Ship Production AI Apps and Agents on Azure
Build, deploy, and operate generative and agentic AI solutions on Azure with Microsoft Foundry — RAG, multi-agent orchestration, vision, speech, and responsible AI for the AI-103 exam.
About This Course
The AI-103, Developing AI Apps and Agents on Azure, is the new associate-level exam that replaces the AI-102 Azure AI Engineer Associate (retiring on 30 June 2026). It earns you the Azure AI Apps and Agents Developer Associate certification, and the shift in scope is significant: where AI-102 was about wiring up Azure AI services, AI-103 is about building agentic AI applications and orchestrating multiple agents on Microsoft Foundry. The difficulty is higher, the scope is broader, and the assumption is that you ship real code.
This is a developer course, end to end. You will plan and secure AI solutions in Foundry, deploy models, implement retrieval-augmented generation, and build agents that call tools, remember context, and collaborate in multi-agent workflows. Then you will operationalize them — observability, evaluations, guardrails, and governance — the parts that separate a demo from production. Every module maps to the official AI-103 skills measured, so nothing on the exam is left uncovered.
You should already be comfortable writing Python and understand the basics of generative AI. If you are coming from AI-102, this is your upgrade path; if you are newer to Azure AI, the AI-901 Azure AI Fundamentals course is a strong on-ramp before you start here.
What You'll Learn
Plan, deploy, secure, and monitor Azure AI solutions built on Microsoft Foundry
Build generative AI apps with RAG, tool-calling, and multistep reasoning
Design and orchestrate single- and multi-agent solutions with safeguards
Implement computer vision, speech, text analysis, and information extraction
Apply responsible AI — guardrails, evaluations, auditing, and agent governance
Prerequisites
- Hands-on experience developing applications with Python
- Familiarity with general AI, generative AI, and core Azure services
- AI-901 (Azure AI Fundamentals) or equivalent knowledge is recommended
Who Is This Course For
This course is for Azure AI engineers, application developers, and machine learning practitioners who build, deploy, and maintain AI solutions on Microsoft Foundry. In this role you collaborate with solution architects, data scientists, DevOps engineers, and cloud security engineers, so the course covers not just building generative and agentic apps but planning, securing, and operating them responsibly.
You should already develop with Python and understand generative AI fundamentals. If you hold or were studying for the AI-102, this is the direct successor and your upgrade path. If you are earlier in your AI journey, start with the AI-901 Azure AI Fundamentals and return here to go deep on shipping production AI apps and agents.
What's Inside
Plan and Manage Azure AI Solutions
- Choosing the right Foundry services, models, and tools for each task
- Designing Azure infrastructure and selecting deployment options
- Managing quotas, scaling, cost, and CI/CD for Foundry projects
- Securing AI with managed identity, private networking, and role policies
Responsible AI for Generative and Agentic Systems
- Safety filters, guardrails, risk detection, and content moderation
- Evaluators, safety evaluations, and explanation tooling
- Auditing with trace logging, provenance metadata, and approval workflows
- Governing agent behavior with oversight modes and tool-access controls
Building Generative AI Applications
- Deploying and consuming LLMs, small, code, and multimodal models
- Implementing retrieval-augmented generation (RAG) in an application
- Designing tool-augmented and multistep reasoning workflows
- Evaluating apps for fabrications, relevance, quality, and safety
Building Agents with Foundry
- Defining agent roles, goals, conversation tracking, and tool schemas
- Integrating retrieval, function-calling, and conversation memory
- Orchestrating multi-agent solutions
- Building autonomous workflows with safeguards and approval-flow controls
Optimizing and Operationalizing AI Systems
- Prompt engineering and tuning model parameters
- Reflection, chain-of-thought evaluation, and self-critique loops
- Observability — tracing, token analytics, safety signals, and latency
- Orchestrating multiple models and hybrid LLM and rules pipelines
Vision, Speech, and Information Extraction
- Image and video generation and editing workflows
- Multimodal understanding, captioning, and visual question answering
- Speech-to-text, text-to-speech, and translation for agentic interactions
- Retrieval, grounding, and document extraction with Content Understanding