Black Friday sale: an extra 40% off every course until Monday, November 30.New Year sale: an extra 30% off every course until Friday, January 1.Memorial Day sale: an extra 25% off every course until Monday, May 31.SysAdmin Day sale: an extra 25% off every course until Sunday, August 1.Back to School sale: an extra 25% off every course until Sunday, August 22.Labor Day sale: an extra 25% off every course until Monday, September 6.Black Friday sale: an extra 40% off every course until Monday, November 29.New Year sale: an extra 30% off every course until Saturday, January 1.Choose your certification →

Copyright (c) 2026 MindMesh Academy. All rights reserved. This content is proprietary and may not be reproduced or distributed without permission.

4.3.2. Creating, Testing, and Publishing Agents

💡 First Principle: Agents are built in Copilot Studio using a no-code/low-code interface — you define the agent's name, purpose, knowledge sources (SharePoint sites, uploaded documents, external URLs), and behavior. Before any user sees it, the agent must be tested to verify it returns accurate responses and stays within its intended scope.

Agent creation steps in Copilot Studio:
  1. Define: Name the agent, write a description, configure its persona (how it should respond)
  2. Add knowledge: Connect to SharePoint sites, upload documents, connect to external APIs
  3. Configure behavior: Write the agent's instructions (its system prompt), set conversation starters, define topics it should handle or avoid
  4. Test: Use the built-in testing interface to validate responses — check accuracy, scope, and edge cases
  5. Publish: Submit for approval (if org-wide), then publish to the chosen channel: Teams, SharePoint, or Microsoft Copilot Chat (formerly Microsoft 365 Copilot Chat)
Knowledge sources an agent can use:
  • SharePoint sites and document libraries
  • Uploaded files (PDFs, Word docs)
  • Public websites (agent can be configured to browse specific URLs)
  • External data via connectors (Power Platform connectors, REST APIs)
  • Dataverse tables

💡 Key Point: An agent is only as good as its knowledge sources. An agent connected to an outdated SharePoint site will give outdated answers. Agents require ongoing maintenance — their knowledge sources must stay current.

Instructions (system prompt): natural-language text that sets the agent's persona, tone, scope and guardrails, for example "answer only questions about warranty terms and politely decline anything else." Instructions, together with which knowledge sources the agent is given, are what keep an agent on topic. Customers can't retrain Microsoft's underlying model, and adding more knowledge sources widens an agent's scope rather than narrowing it.

Topics, actions and generative answers:
  • A topic is an authored conversation path started by trigger phrases. Built-in system topics include Fallback (nothing matched, or no answer was found) and Escalate (hand the conversation to a live agent or contact center).
  • An action (tool) does work in another system: it calls an API, runs a Power Automate flow or uses a connector. Actions are how an agent gets live data such as the current status of an order; a knowledge source supplies reference content to ground answers.
  • Generative answers let the AI compose a response from the connected knowledge sources without a pre-written topic, while authored topics follow scripted paths.
Keeping an agent current and diagnosing problems:
  • The test pane runs the draft you're editing. Users in a channel such as Teams see only what has been published to a channel that has been added, and changes to knowledge, instructions or topics reach them only after the agent is published again.
  • SharePoint knowledge is found through Microsoft 365 search, so an edited document shows up in answers only after the search index has picked up the change. That lag clears by itself and needs no rebuild.
  • When an agent says it has no information on a subject, that content isn't in its knowledge sources, so add it.
  • When an action returns errors, test that action on its own in Copilot Studio to check its connection and credentials before changing anything else.

⚠️ Exam Trap: Copilot Studio is the tool for creating agents. It is a separate product from the Microsoft 365 admin center, and it uses Power Platform infrastructure under the hood. This is why agent monitoring is done in both the M365 admin center (for Copilot-side metrics) and the Power Platform admin center (for agent-side metrics).

Reflection Question: A user builds an agent connected to the company's public-facing FAQ SharePoint site. After publishing, users report that the agent gives correct answers about current products but wrong answers about prices. What is the most likely root cause, and what maintenance action is needed?

See how it connects
Alvin Varughese
Written byAlvin Varughese
Founder•26 professional certifications earned