AI Capability Pathway: Understand, Use, Build and Govern

AI literacy for nontechnical participants.

Level prerequisites

• None.
• No technical or programming knowledge is required.

Level learning goals

• Explain AI and generative AI in everyday language.
• Understand broadly how modern AI systems produce responses.
• Distinguish AI capabilities from common misconceptions.
• Recognize major benefits and limitations of AI.
• Identify basic privacy, security and reliability risks.
• Understand why human oversight remains important.
• Participate more confidently in organizational discussions about AI.

Curriculum

Understanding Artificial Intelligence

• What artificial intelligence means
• Traditional software versus AI systems
• Machine learning and neural networks
• Generative AI
• Large language models
• Multimodal AI
• AI assistants, copilots and agents

How Modern Generative AI Works

• Training and model knowledge
• Tokens and language prediction
• Prompts and responses
• Context
• Why AI appears intelligent
• Model knowledge, uncertainty and the limits of generated answers

Demystifying AI

• Intelligence versus prediction
• AI reasoning capabilities
• Creativity and generation
• Common AI myths
• Current limitations

Opportunities Created by AI

• Productivity
• Knowledge work
• Research and communication
• Automation
• Decision support
• Emerging workplace applications

Understanding AI Risk

• Hallucination
• Bias
• Privacy
• Confidential information
• Misinformation
• Cybersecurity misuse
• Employment and societal impact
• Overreliance on AI

The Wider AI Safety Discussion

• Geoffrey Hinton and modern AI development
• Near-term AI risks
• Long-term AI concerns
• Increasing model capabilities
• Human control and oversight
• Responsible adoption

Becoming AI Aware

• Questioning AI output
• Protecting organizational information
• Knowing when AI should not be used
• Human accountability
• Continuous AI literacy

Practical prompting, context, review and responsible departmental use.

Level prerequisites

• Level 1 or equivalent basic AI awareness.
• General experience using workplace applications.

Level learning goals

• Use generative AI more effectively in daily work.
• Construct clear and structured prompts.
• Understand context windows and context limitations.
• Apply context-engineering techniques.
• Validate and refine AI-generated output.
• Recognize confidential and inappropriate information.
• Identify useful AI opportunities within business departments.
• Understand personal and departmental AI assistants.
• Apply responsible-use and governance principles.
• Recognize prompt injection and other common AI threats.

Curriculum

Working Effectively with Generative AI

• AI conversation fundamentals
• Prompts and instructions
• Tasks, objectives and constraints
• Output requirements
• Iterative interaction
• Verification and refinement

Prompt Engineering

• Prompt structure
• Roles and perspectives
• Context
• Constraints
• Output formatting
• Prompt refinement
• Reusable prompt patterns

Context Engineering

• Understanding context windows
• Relevant versus irrelevant context
• Context quality
• Instruction hierarchy
• Document context
• Conversation context
• Context limitations

Improving AI Reliability

• Recognising and investigating possible hallucinations
• Verification
• Source checking
• Assumptions
• Ambiguous instructions
• Human review

AI for Business Productivity

• Human Resources
• Finance
• Operations
• Administration
• Sales
• Customer service
• Management
• Research and documentation

Working with Organizational Information

• Public information
• Internal information
• Confidential information
• Personal information
• Regulated information
• Intellectual property

Responsible AI Usage

• Bias
• Privacy
• Security
• Accuracy
• Accountability
• Automation bias
• Appropriate human oversight

AI Governance for Users

• Organizational AI policies
• Acceptable-use requirements
• Approved versus unapproved AI tools
• Shadow AI
• Data classification
• Auditability
• Compliance responsibilities

Personal and Departmental AI Assistants

• Personal AI assistants
• Department-specific assistants
• Instructions and boundaries
• Knowledge sources
• Permissions
• Human escalation

AI Security Awareness

• Prompt injection
• Indirect prompt injection
• Data leakage
• Malicious content
• Unsafe instructions
• Excessive trust
• Credential and secret exposure

Selected prepared implementation examples; not a complete advanced engineering programme.

Level prerequisites

• Level 2 or equivalent generative AI knowledge.
• General IT knowledge.
• Familiarity with Microsoft 365 or Power Platform is beneficial.
• Basic knowledge of APIs is beneficial.

Level learning goals

• Understand the components of Microsoft's enterprise AI stack.
• Create and configure a basic Copilot Studio agent using a prepared example
• Integrate agents with business actions and workflows.
• Understand Power Automate and agent-flow architectures.
• Consume AI capabilities through APIs.
• Understand Microsoft Foundry and model endpoints.
• Explain RAG architecture and enterprise grounding.
• Understand authentication and permission requirements.
• Compare different approaches to enterprise agent development.
• Apply basic security and governance controls.

Curriculum

Enterprise Generative AI Architecture

• Models
• Prompts
• Context
• Knowledge
• Tools
• Actions
• Agents
• Enterprise systems

Microsoft Enterprise AI Ecosystem

• Microsoft 365 Copilot
• Copilot Studio
• Power Automate
• Microsoft Foundry: selected current platform capabilities
• Azure AI/Foundry services: current naming and availability
• Azure AI Search
• Microsoft Entra ID
• Microsoft Purview

Copilot Studio Fundamentals

• Agent creation
• Agent instructions
• Knowledge sources
• Topics
• Generative orchestration
• Tools and actions
• Connectors
• Testing and publishing

Agent Flows and Power Automate

• Agent flows
• Power Automate cloud flows
• Workflow triggers
• Business-system connectors
• Agent actions
• Human approvals
• Returning results to agents

API-Based AI Integration

• REST API concepts
• AI endpoints
• Authentication
• HTTP actions
• Request and response handling
• Enterprise API integration

Microsoft Foundry

• Models and deployments
• Model endpoints
• Foundry projects
• Foundry Agent Service
• Tools and enterprise integrations
• Identity and access

Retrieval-Augmented Generation

• RAG concepts
• Enterprise knowledge
• Document ingestion
• Indexing
• Azure AI Search
• Vector and hybrid retrieval
• Grounded responses
• Citations

Selecting an Agent Architecture

• Copilot Studio agents
• Agent flows
• Power Automate
• API-driven solutions
• Foundry agents
• Architecture selection

Basic Agent Governance

• Authentication
• Authorization
• Least privilege
• Data loss prevention
• Connector controls
• Knowledge boundaries
• Human approval
• Logging and auditing

A separately scoped advanced technical track; duration determined separately.

Level prerequisites

• Level 3 or equivalent enterprise AI implementation knowledge.
• Familiarity with cloud architecture.
• Understanding of APIs and authentication.
• Power Platform or Azure experience is strongly recommended.
• Basic software-development lifecycle knowledge is recommended.

Level learning goals

• Design production-oriented enterprise AI architectures.
• Apply AI SDLC and lifecycle-management principles.
• Understand MLOps and GenAIOps practices.
• Engineer and evaluate enterprise RAG systems.
• Design secure tool-using agents.
• Apply identity and least-privilege architectures.
• Monitor AI quality, safety, usage, latency and cost.
• Manage prompt, model and knowledge changes.
• Understand agentic retrieval and advanced RAG.
• Design multi-agent architectures.
• Implement human-in-the-loop and approval controls.
• Establish production governance for enterprise AI.

Curriculum

AI Solution Architecture

• Enterprise AI architecture patterns
• Model layer
• Orchestration layer
• Knowledge layer
• Tool layer
• Identity layer
• Governance layer
• Observability layer

AI Software Development Lifecycle

• Requirements and risk assessment
• Design
• Development
• Evaluation
• Security testing
• Deployment
• Monitoring
• Continuous improvement
• Retirement

MLOps and GenAIOps

• Model lifecycle
• Prompt lifecycle
• Knowledge lifecycle
• Configuration management
• Versioning
• Environment promotion
• CI/CD concepts
• Rollback strategies

AI Change Management

• Model changes
• Prompt changes
• Knowledge changes
• Tool changes
• Connector changes
• Permission changes
• Regression testing
• Production approvals

Advanced RAG Engineering

• Chunking strategies
• Embeddings
• Vector retrieval
• Hybrid retrieval
• Metadata filtering
• Reranking
• Query transformation
• Agentic retrieval and version-specific capabilities, including preview query planning where applicable
• Knowledge-base architecture
• Permission-aware retrieval

RAG Governance and Security

• Source authorization
• Document permissions
• Sensitive information
• Data leakage
• Index security
• Knowledge freshness
• Assessing groundedness; retrieval does not eliminate incorrect answers
• Retrieval auditing

Advanced Agent Engineering

• Tool-using agents
• Planning and reasoning
• Structured tool invocation
• API tools
• MCP-based integration and explicit identity/action boundaries
• Agent memory
• State management
• Action boundaries

Agent Security

• Direct prompt injection
• Indirect prompt injection
• Tool poisoning
• Data exfiltration
• Excessive agency
• Privilege escalation
• Identity propagation
• Credential management

Multi-Agent Systems

• Specialized agents
• Supervisor agents
• Delegation
• Task routing
• Agent-to-agent communication
• Shared knowledge
• Shared state
• Permission boundaries
• Failure management

Deterministic Automation versus Agentic Automation

• Traditional workflows
• AI-assisted workflows
• Autonomous agents
• Agent flows
• Power Automate
• Copilot Studio
• Foundry agents
• Architecture trade-offs

AI Evaluation

• Response quality
• Groundedness
• Relevance
• Safety
• Retrieval quality
• Tool-selection accuracy
• Agent task completion
• Regression evaluation

AI Observability and Monitoring

• Usage telemetry
• Agent activity
• Model consumption
• Token consumption
• Cost monitoring
• Latency
• Failure rates
• Quality degradation
• Security events

Enterprise AI Governance

• Environment strategy
• Identity and RBAC
• Data loss prevention
• Connector governance
• Microsoft Purview
• Microsoft Entra
• Audit requirements
• Risk classification
• Production controls

Production AI Operations

• Development-to-production promotion
• Testing gates
• Release management
• Incident management
• Rollback
• Usage governance
• Service ownership
• Continuous evaluation