AI-103 Accelerator

Building AI Apps and Agents on Azure - 3 days

Microsoft’s new AI-103 direction matters because it reflects a real shift in how AI systems are being built: away from isolated service demos and toward integrated applications, agent workflows, model evaluation, multimodal processing, and production deployment on Microsoft Foundry and Azure. Microsoft Learn lists AI-103T00: Develop AI apps and agents on Azure as a new intermediate course for AI engineers and developers, officially available from March 31, 2026, with a standard duration of 4 days. Its stated focus is building AI-infused applications with Microsoft Foundry, including generative AI apps, AI agents, tool and knowledge connections, multimodal capabilities, and complex-content understanding.

This 3-day outline is designed as a compressed, industry-style version of that curriculum rather than an academic survey. It also accounts for Microsoft’s transition away from AI-102. Microsoft lists AI-102T00 as retiring on April 30, 2026, with AI-103T00 named as its replacement course; separately, Microsoft lists the AI-102 certification exam as retiring on June 30, 2026. Microsoft’s March 2026 certification announcement further explains that AI-103 aligns to newer generative and agentic architectures, with beta timing in April 2026 and general availability expected in June 2026. Community replies from Microsoft Learn experts also indicate that AI-102 and AI-103 are not equivalent, and there is no automatic transition from the old certification to the new one.

The instructor for this program brings over 30 years of industry experience and will teach the material through current, employer-relevant practice: what teams are actually building now with Azure AI Foundry, model catalogs, evaluation workflows, agent frameworks, orchestration, and multimodal pipelines, rather than treating AI as a purely theoretical subject.

Learning outcomes

Plan and structure AI application development on Azure using Microsoft Foundry and related Azure services.
Build generative AI applications with deployed models, SDKs, prompts, and application integration patterns.
Develop AI agents with Microsoft Foundry Agent Service and Microsoft Agent Framework.
Extend agents with tools, functions, knowledge connections, and workflow orchestration patterns.
Evaluate generative AI behavior and apply responsible AI practices in Azure AI Foundry.
Design solutions that incorporate multimodal content and complex-content understanding.
Recognize where legacy AI-102-era skills still matter and where the new AI-103 emphasis changes architecture and delivery choices.

Prerequisites

Working knowledge of Python. Microsoft positions AI-103 for developers already familiar with Python and APIs/SDKs on Azure.
Basic familiarity with Azure and the Azure portal.
Comfort using APIs, SDKs, and application integration patterns.
Foundational understanding of generative AI concepts.
Helpful but not mandatory: prior exposure to AI-102 topics such as Azure AI services, language, vision, or information extraction.

Detailed training outline

Course positioning and Azure AI landscape
1. The Microsoft AI certification and course transition
  1. What AI-102 covered
  2. Why Microsoft introduced AI-103
  3. Courseware retirement vs exam retirement
  4. Implications for teams, learners, and certification paths
2. The modern Azure AI application stack
  1. Microsoft Foundry and Azure AI services
  2. Foundry projects, resources, and environments
  3. Models, agents, tools, and knowledge as first-class building blocks
3. Delivery mindset for production AI
  1. From prototype to deployable application
  2. Developer responsibilities in AI solution lifecycles
  3. Responsible AI as an engineering concern, not a compliance afterthought
Planning and preparing to develop AI solutions on Azure
1. Solution architecture fundamentals
  1. Selecting the right Azure AI components
  2. Matching use cases to models, tools, and orchestration styles
  3. Choosing between direct app patterns and agent patterns
2. Environment and resource preparation
  1. Azure subscriptions, resource groups, and access setup
  2. Foundry workspaces, projects, and resource organization
  3. Model access, deployment readiness, and configuration baselines
3. Security and governance foundations
  1. Identity, secrets, and access boundaries
  2. Development vs test vs production considerations
  3. Governance checkpoints for AI workloads
4. Application lifecycle planning
  1. Prototyping, iteration, evaluation, and release flow
  2. Traceability and reproducibility expectations
  3. Operational readiness criteria
Building generative AI applications in Azure
1. Generative AI application fundamentals
  1. Core prompt-response application patterns
  2. System instructions, user prompts, and output shaping
  3. Stateless vs stateful interaction design
2. Working with models in Azure AI Foundry
  1. Model catalog concepts
  2. Model selection criteria
  3. Deployment configuration and inference access
3. Developing with the Azure AI Foundry SDK
  1. SDK structure and project setup
  2. Connecting applications to deployed models
  3. Managing requests, responses, and runtime settings
4. Prompt and interaction design
  1. Instruction design patterns
  2. Output constraints and formatting control
  3. Grounding strategies for better reliability
5. Integrating generative features into applications
  1. Service-layer integration
  2. Web and API integration patterns
  3. Handling latency, retries, and failure modes
6. Responsible generative AI implementation
  1. Safety-by-design principles
  2. Content risk awareness
  3. Human oversight and escalation design
Evaluating and improving generative AI performance
1. Why evaluation matters
  1. Functional quality vs model quality
  2. Reliability, relevance, and consistency
  3. Evaluation in iterative delivery cycles
2. Evaluation workflows in Azure AI Foundry
  1. Baseline creation
  2. Test dataset setup
  3. Comparative evaluation approaches
3. Assessment dimensions
  1. Accuracy and helpfulness
  2. Groundedness and hallucination risk
  3. Safety and policy alignment
  4. Output consistency and task completion
4. Tuning and improvement loops
  1. Prompt refinement
  2. Model substitution and comparison
  3. Tool-use adjustments
  4. Knowledge-source refinement
5. Release readiness
  1. Acceptance thresholds
  2. Regression checks
  3. Monitoring expectations after deployment
AI agent fundamentals on Azure
1. What makes an AI agent different from a standard AI app
  1. Goal-directed behavior
  2. Tool use and action-taking
  3. Multi-step reasoning and workflow execution
2. Agent use-case identification
  1. When an agent is appropriate
  2. When a simple app pattern is better
  3. Boundaries of autonomy in enterprise systems
3. Microsoft Foundry Agent Service concepts
  1. Agent architecture
  2. Instructions, memory, tools, and knowledge connections
  3. Runtime interaction model
4. Microsoft Agent Framework concepts
  1. SDK role in application development
  2. Agent composition principles
  3. Framework-level orchestration patterns
Developing AI agents with Microsoft Foundry and Visual Studio Code
1. Agent development workflow
  1. Project scaffolding
  2. Agent configuration
  3. Development toolchain setup
2. Building the first agent
  1. Defining purpose and behavior
  2. Configuring prompts and instructions
  3. Establishing execution boundaries
3. Extending agent capabilities
  1. Tool registration
  2. Function calling patterns
  3. Connecting business logic and APIs
4. Testing and debugging agents
  1. Playground-based testing
  2. Prompt and tool debugging
  3. Behavioral validation across scenarios
5. Deploying and integrating agents
  1. Deployment packaging
  2. Application integration
  3. Operational verification
Developing agents with Microsoft Agent Framework
1. Framework architecture
  1. Core abstractions
  2. Agent definitions and execution flow
  3. Tool and connector integration
2. Agent creation patterns
  1. Single-purpose agents
  2. Task-oriented agents
  3. Knowledge-enabled agents
3. Tool augmentation
  1. Calling external functions
  2. Structuring actions and responses
  3. Managing tool-result feedback loops
4. Enterprise integration
  1. Linking agents to internal services
  2. Designing controlled action paths
  3. Reliability and maintainability considerations
Orchestrating multi-agent solutions
1. Multi-agent architecture fundamentals
  1. Coordinator and specialist patterns
  2. Delegation and handoff
  3. Conversation and state boundaries
2. Orchestration strategies
  1. Sequential workflows
  2. Parallel agent collaboration
  3. Escalation and fallback paths
3. Workflow composition in Microsoft Foundry
  1. Agent-driven workflows
  2. Task decomposition
  3. Result aggregation
4. Governance for multi-agent systems
  1. Control points
  2. Observability needs
  3. Risk management in distributed reasoning flows
Knowledge connections and retrieval-enabled applications
1. Why grounding matters
  1. Enterprise knowledge challenges
  2. Reducing hallucination in production scenarios
  3. Retrieval as part of application design
2. Knowledge-source integration patterns
  1. Document collections
  2. Structured and semi-structured content
  3. External enterprise knowledge endpoints
3. Designing retrieval-enhanced responses
  1. Query formulation
  2. Context injection
  3. Citation and answer traceability patterns
4. Tool vs knowledge decisions
  1. When an agent should retrieve
  2. When an agent should act
  3. Combining retrieval and action safely
Multimodal AI capabilities
1. Multimodal solution landscape
  1. Text, image, audio, video, and document understanding
  2. Choosing multimodal patterns for business use cases
2. Using multimodal models in application flows
  1. Input handling and normalization
  2. Cross-modal reasoning considerations
  3. Output design for downstream systems
3. Complex-content understanding
  1. Documents with mixed structure
  2. Rich media processing patterns
  3. Extracting structured insight from varied inputs
4. Multimodal pipeline design
  1. Ingestion
  2. Processing
  3. Enrichment
  4. Application consumption
Complex-content understanding and content extraction
1. Understanding unstructured and semi-structured content
  1. Documents
  2. Forms
  3. Reports
  4. Media-derived artifacts
2. Content Understanding capabilities
  1. Schema-driven extraction
  2. Structured output generation
  3. Workflow integration
3. Application design for extracted insights
  1. Post-processing and validation
  2. Routing to downstream systems
  3. Human review patterns where needed
Applied app-and-agent engineering practices
1. Design patterns for real-world delivery
  1. Internal productivity assistants
  2. Knowledge assistants
  3. Action-oriented process agents
  4. Human-in-the-loop operational assistants
2. Performance and scalability considerations
  1. Throughput and concurrency
  2. Model and tool invocation cost awareness
  3. Response-time optimization
3. Reliability engineering
  1. Timeout handling
  2. Retry and fallback logic
  3. Partial-failure design
4. Observability and operations
  1. Logging and tracing
  2. Evaluation in production
  3. Change management for prompts, tools, and models
Responsible AI and production governance
1. Responsible AI principles in the delivery lifecycle
  1. Fairness, reliability, privacy, transparency, and accountability
2. Safety controls for generative and agentic systems
  1. Input screening
  2. Output moderation
  3. Access control and human approval gates
3. Compliance-aware solution design
  1. Data handling considerations
  2. Enterprise governance checkpoints
  3. Documentation and auditability
4. Production readiness review
  1. Technical readiness
  2. Operational readiness
  3. Risk readiness
Capstone architecture synthesis
1. End-to-end solution framing
  1. Defining the application or agent scenario
  2. Mapping business goals to Azure components
  3. Selecting app, agent, or multi-agent patterns
2. Delivery blueprint
  1. Resource layout
  2. Development workflow
  3. Evaluation workflow
  4. Deployment workflow
3. Transition to implementation
  1. Backlog shaping
  2. Team roles
  3. Next-step lab or project planning

Practical, connected learning

My wider training approach brings hands-on implementation and systems thinking together, connecting technology with real operational needs.

AI-103 Accelerator