Copilot Agents: build a custom assistant for your role.
Copilot Agents are the next step up from Pages — purpose-built AI assistants for specific roles or workflows. This is Pro+ territory because the value is enterprise-grade: think of an HR policy agent, a sales-research agent, an IT helpdesk agent. This lesson walks the full build of three real agents from scratch.
The mental model
Agents are Custom GPTs for the Microsoft world.
Like ChatGPT's Custom GPTs but built for your organization — connected to your SharePoint, your Teams chats, your company knowledge. The agent knows your context; you don't have to re-explain it every time.
Workflow 01 Build a sales-research agent
Agent: "Customer Prep"
Build an agent that researches any prospect before a meeting — pulling from your CRM, past emails, and the web.
The prompt that works
Best use cases
- Sales reps prepping for calls
- Account managers in QBRs
- Customer success teams onboarding new accounts
Workflow 02 Build an HR policy agent
Agent: "HR Help"
Most HR questions are repetitive: PTO policy, benefits enrollment, expense rules. Build an agent that answers these from your actual policy docs.
The prompt that works
Best use cases
- Mid-size company HR teams overwhelmed with repetitive questions
- Onboarding flows for new hires
- Manager self-service for policy questions
Workflow 03 Build an IT helpdesk agent
Agent: "IT Triage"
First-line IT support is mostly password resets, VPN issues, and how-to questions. An agent handles 60-70% of tickets before they reach a human.
The prompt that works
Best use cases
- Internal IT teams at companies with high ticket volume
- Remote-first companies where in-person help isn't available
- After-hours triage when IT is offline
Final challenge: build your first real agent
Pick the repetitive question you get asked most often at work. Build a Copilot Agent that handles it:
- Open Copilot Studio
- Name the agent (something descriptive)
- Connect it to the relevant data source (SharePoint folder, Teams channel, knowledge base)
- Write 5-10 sentences of instructions that constrain its behavior
- Test with 5 real questions you'd expect
- Iterate the instructions until it answers well
- Deploy to a small group, refine based on feedback
Time invested: 1-2 hours. Time saved: ongoing forever.
What you can do now
- Understand when an Agent beats a one-off Copilot conversation
- Build agents that connect to real organizational data (SharePoint, Teams, CRM)
- Write agent instructions that constrain hallucination and add citations
- Configure graceful escalation paths for questions agents can't answer
- Pick the right repetitive workflow to automate first