Skip to content

Phase 3: Build Agents

Goal: Build department-specific agents that automate real business processes. This is the transform.

  • Identify 2-3 high-impact business processes per department
  • Map processes to agent types: retrieval agents (Agent Builder), custom agents (Copilot Studio), autonomous agents
  • Define success metrics and ROI targets

3.2 Build & Deploy Agents (Copilot Studio)

Section titled “3.2 Build & Deploy Agents (Copilot Studio)”

Following the LUCSP lab progression:

Level 1 — Retrieval Agent (Agent Builder)

  • Build a simple retrieval agent using Copilot Studio Agent Builder
  • Quick win to demonstrate value
  • ~30 minutes to build

Level 2 — Custom Agent (Copilot Studio Full)

  • Copilot Studio UI: create agent, extend knowledge base, configure generative AI orchestration
  • Define Topics: single-turn, multi-turn, adaptive cards, multi-language support
  • Integrate Tools: Excel, Dataverse, Power Apps tables
  • Build Power Automate flows for process automation
  • Example: Customer Service Virtual Agent for e-commerce (product discovery, appointment booking, returns/refunds)

Level 3 — Advanced / Autonomous Agents

  • Autonomous agent capabilities with triggers (e.g., email-triggered automation)
  • MCP (Model Context Protocol) server connectivity for third-party integration
  • Multi-agent architectures and orchestration
  • Computer Use controls, Code Interpreter, Document Processor
  • Advanced approval workflows
  • Configure Copilot Studio access with Conditional Access
  • Manage publishing approvals and agent permissions (Zero Trust principles)
  • Apply DLP to limit sensitive data in agent interactions
  • Restrict agent access to approved data sources only
  • Enable audit logging and monitoring for all agent activity
  • Use Agent 365 for unified agent inventory and policy management
  • Set up cost controls and consumption management (pay-as-you-go)
  • Leverage Work IQ for contextual insights: data (Graph + LOB), memory (conversational + explicit), inference (predictive actions)
  • Position Work IQ-powered agents: Workforce Insights Agent (leadership analytics), People Agent (colleague discovery), Learning Agent (personalized microlearning)
  • Activities: Agent strategy workshop, agent build sprints, security hardening, adoption
  • Deliverable: Working agents in production, governance framework, ROI report
  • Exit Criteria: Agents deployed and active, measurable business impact