Advanced Copilot Course Boston October 1, 2026

Moving from AI Prompts to AI Agents

• Going beyond the limits of prompt-based AI
• Identifying tasks that fail with AI prompts but are possible with agents
• Expanding AI into infrastructure
• What makes an AI system an “agent”
• Stateless vs stateful AI behavior
• Autonomous workflows vs. human-in-the-loop
• When to use an agent and when to avoid agents

Understanding Copilot AI Agent Types and Patterns

• AI Agent types
  • Informational agents
  • Decision-support agents
  • Review & compliance agents
  • Orchestration agents
  • Intake & triage agents
• Using a pattern library to build agent behavior
  • Producing
  • Summarizing
  • Escalating
  • Recommending
• Mapping business tasks to an agent pattern

Custom AI Agents in Microsoft Copilot

• Creating agents that perform with high-confidence
• Capabilities of Copilot vs Copilot Studio vs Power Automate
• Knowledge grounding and retrieval
• Security boundaries and permissions
• Instruction hierarchy and structure
• Why AI hallucinates and how you can use agents to reduce this risk

Defining Copilot AI Agent Purpose, Scope, and Guardrails

• Creating the foundation of an agent: writing an agent charter
• Controlling the scope of what an agent can and cannot do
• Developing guardrails for allowed sources, forbidden actions, and escalation conditions
• Creating rules for output consistency 
• Embedding Refusal behavior (when agents should say “I don’t know”)
• Completed Agent Design Canvas (provided template)

Knowledge Grounding Strategies:

• Agents are only as good as their knowledge
• Safely using online content repositories
• Curating document sets
• Version control and document drift
• Handling conflicting sources
• Structured vs unstructured knowledge
• Designing a clean knowledge set for an agent

Building a Custom AI Agent in Copilot Studio

• Copilot Studio overview
• Creating agent instructions
• Connecting knowledge sources
• Testing agent responses
• Making iterative improvements
• Building a working agent grounded in knowledge
• Testing agents with controlled scenarios

Advanced AI Agent Behaviors: Making an Intelligent AI agent

• Multi-step reasoning
• Developing clarifying questions
• Enforcing structured outputs
• Applying self-review and correction
• Creating conditional logic without code
• Applying thresholding and logic
• Adding validation logic

AI Agents in Real Business Workflows

• Moving from from isolated tools to real systems
• Triggering agents via:
  • Email
  • File upload
  • Forms
• Creating human approval loops
• Using logging and traceability
• Determining when to introduce automation
• Developing an automation process

Governance, Risk, and Enterprise Controls for Copilot AI Agents

• Creating systems for public use
• Preventing prompt drift
• Defining access control and permissions
• Auditing agent outputs
• Sensitive data handling
• Addressing bias and compliance concerns
• Sunsetting and updating agents

Scaling AI Agents Across an Organization

• Turn pilot projects into platforms
• Agent libraries and reuse
• Role-based agent frameworks
• Internal enablement models
• Training others responsibly
• Change management strategies

Measuring Success and ROI of Copilot AI Agents

• Proving value and determining productivity metrics
• Determining quality and consistency gains
• Evaluating risk reduction
• Continuous improvement cycles
• Defining KPIs for your agent and OKRs for AI adoption

Building your own Copilot AI agent

• Design and build a complete custom agent
• Defining your agent
  • Purpose
  • Knowledge
  • Guardrails
  • Success metrics
• Testing your AI agent
• Design decisions and outcomes review