Make your Own Agents in Copilot | Complete Tutorial

What “agents” are (core idea)

• Agents expand AI tools by acting as customized assistants for specific jobs/purposes.
• They can follow preset specifications/rules and run workflows across multiple applications (though the tutorial’s “quick build” focuses on limited integration).
• Agents can be configured to work independently or as part of multi-step flows (more advanced in Copilot Studio).

Where agents are available + licensing constraints

• The course focuses specifically on Microsoft Copilot (not Gemini/Claude/etc.).
• Free Copilot access: agents are not available (or are extremely limited).
• Microsoft 365 Copilot (business/enterprise): agents and agent-building features require organization subscriptions.
• Some advanced agents/features (e.g., certain enterprise search/work mode capabilities, researcher/analyst) require the full Microsoft 365 Copilot license.

Course structure + learning goals (11 sections)

The speaker outlines 3 main goals:

1. Explain what agents are and why they help (time saved, better focus/results).
2. Show how to use existing Microsoft/third-party agents inside Copilot and Microsoft Teams.
3. Teach how to build your own agents (mainly using Copilot Studio and Microsoft’s simpler Copilot Agent Builder).

---

Using pre-built agents in Copilot (review/guide style workflow)

Example: “Career Coach” agent (Microsoft-built)

• Added from the Copilot Agents sidebar category (All agents → Microsoft agents).
• The tutorial demonstrates:
  • Adding an agent via an Add button.
  • Switching chat modes between standard Copilot and the selected agent.
  • Asking a task prompt and receiving guided questions + specialized research.
• Note: some users may be prompted to upgrade depending on subscription.

Invoking agents inside an existing chat

• Type an “@” / add-symbol + agent name to invoke an already-installed agent without starting a new chat.

Removing agents

• Agents can be uninstalled from the left sidebar via a three-dots menu.

---

Using agents in Microsoft Teams

• Agents are discovered via Teams sidebar Apps / Get more apps.
• The tutorial notes agents may appear differently:
  • Some create a dedicated chat
  • Others appear as an extra sidebar button
• Example: “Idea Coach” agent is added, then opened via Open with Copilot.

---

Agent parameters + “enterprise search” capability check

Recommended “agent parameters” prompt

The tutorial recommends asking the agent to:

• Describe parameters, including capabilities/functions
• List rules/constraints
• Explain technical configuration

Enterprise search / work mode capability

• Agents may use enterprise search / work mode to access organization data (email/messages/shared docs),
  • typically only under the full license.

---

Two “spotlight” agents: Researcher vs Analyst (analysis + tradeoffs)

Researcher agent

• Purpose: deep research with extra time to compare resources and reduce inaccuracies.
• Demonstrated behaviors:
  • Asks follow-up questions
  • Produces citations
  • Uses company docs (e.g., an employee manual)
• Tradeoff explicitly noted:
  • Cannot generate images/pictures (as a rule/constraint)

Analyst agent

• Purpose: insights from actual data (more analytics-oriented).
• Demonstrated behavior:
  • Works well for decisions based on spreadsheets and market trends
  • Identifies missing data
  • Can offer charts or scenario simulation
• Tradeoff:
  • Takes longer to respond
  • Optimized so users don’t need to write all configuration prompts themselves

---

Third-party agents (product integration + security)

Example: Canva agent

• Added from Copilot third-party agents list.
• The tutorial demonstrates:
  • Asking the agent for parameters/capabilities (design creation, retrieving designs, style suggestions)
  • Referencing organization content (e.g., using an email thread via file/email lookup syntax)
  • Requesting an Instagram promotional graphic
• Security/integration steps:
  • Requires allowing a connection to the third-party service
  • May require sign-in to Canva to edit/download final outputs
  • Supports iterative follow-ups (generate designs with specific date; preview/edit)

Controlling permissions

• Connected apps can be reset:
  • Copilot settings → connected Agents/App section → Reset (revokes access)
• Agents can be fully removed via uninstall.

---

Excel “agent mode” (specialized operational agent)

• Excel agent mode is an Excel-only tool that can take actions the regular Copilot chat cannot.
• Demonstrated workflow:
  • Regular Copilot answers in chat but typically does not edit the sheet
  • Agent mode can:
    • sort/filter data
    • apply conditional formatting (even when not directly requested, based on natural-language intent)
    • calculate derived fields (e.g., profit margin)
    • generate dashboards (tables/graphs) and provide insights/advice
• Key limitations noted:
  • It “cannot make up information” or generate original content like Word drafting
  • It relies on analysis/calculations and can project trends

---

Building your own agents (agent builder + Copilot Studio)

1) Copilot “Agent Builder” (simpler, in-Copilot tool)

Main steps

• Go to Copilot website → Agent section → New agent
• Use Describe vs Configure tabs to generate instructions and then refine
• Provide:
  • Name + description
  • Instructions (role, rules, constraints, “don’t guess,” source checking)
  • Knowledge sources (documents/web)
  • optional abilities: generate docs/code/images (capability toggles)
  • suggested prompts for the agent UI
• Test with a preview/test panel, then Create

Knowledge source privacy/usability tradeoff

• If you attach documents from OneDrive/SharePoint, teammates can use them only if files are shared appropriately.
• If you upload files from your computer, they’re bundled with the agent so all users can access—this may increase privacy/data exposure concerns.

Testing + iteration requirement (explicit emphasis)

• The tutorial stresses:
  • testing multiple prompts
  • updating instructions/knowledge
  • revising until confident

Sharing and feature/license limits

• Agents can be shared within the organization via share links.
• If the builder used features requiring the full Copilot license, recipients on baseline accounts may see an upgrade requirement message.

When to use Copilot Studio instead

• If Agent Builder can’t meet needs (advanced autonomous actions, integrations), use Copilot Studio.
• Copilot Studio may require admin involvement for licensing, usage fees, and deployment.

---

2) Copilot Studio (more advanced agent building)

Setup and building flow

• Start at copilotstudio.microsoft.com
• Use an AI assistant to generate an agent from a description or create a blank agent/template
• Key configurations shown:
  • choose the AI model powering the agent (and model strength guidance)
  • define triggers (for autonomous actions/events; highlighted but not deeply covered)
  • write instructions (role, expectations, limits)
  • set knowledge sources (e.g., employee manual)
  • allow/deny public internet research
  • add suggested prompts

Testing inside Copilot Studio

• Preview panel shows:
  • responses plus citations tied to knowledge sources
• Emphasizes multiple rounds of testing and editing before finishing

“Topics” to force scenario-specific behavior

• Topics are conversation control structures triggered by user input.
• Demonstrated:
  • Simple trigger: if user includes “thanks/thank you,” respond “you’re welcome”
  • More complex: a fallback topic can branch via conditions
• Topics can include:
  • trigger phrases/recognition logic
  • conversation node trees
  • optional code editor/YAML approach (advanced)

---

Publishing and deploying agents (auth, channels, approval, costs)

Publishing inside the organization (Microsoft authentication)

• Publishing requires more than clicking “publish”:
  • configure authentication (who can access; Microsoft identity is easiest)
  • then publish to chosen channels
• Channels demonstrated:
  • Microsoft 365 Copilot
  • Microsoft Teams
• Deployment steps for the org:
  • configure availability options (e.g., “show everyone in my org”)
  • submit to administrator for approval
  • teammates can add from an agent store in Copilot/Teams

External publishing (web / third-party services)

• External users require reconsidering authentication:
  • either require authentication (more secure), or set no authentication for public web demo scenarios
• If no authentication is enabled, the tutorial warns:
  • knowledge sources may become publicly accessible
  • some features may not work
  • you must ensure sensitive information isn’t exposed

External channels list + “demo website” publishing

• Channels like Slack/WhatsApp/other services require more setup and generally require Copilot Studio licensing + credits for external users.
• The tutorial also demonstrates a simpler demo website channel:
  • set a welcome message + conversation starters
  • copy/share a link
  • verify responses from built-in knowledge sources
• Note: demo website is easiest but limited/not ideal for long-term use.

Cost/credit considerations (usage credits)

• Agents can consume paid resources (usage credits).
• Credits may be needed for:
  • baseline users to use advanced agents
  • autonomous actions / program-trigger actions
  • external deployments
• Credits can be purchased prepaid or pay-as-you-go (pricing/estimation depends on usage).

---

Main speakers/sources

• Speaker: Nick — host/trainer introducing and demonstrating Copilot/Microsoft 365 agent capabilities and the tutorial steps.