Class Introduction
This was an AI Agent creation class led by Dinesh where students learned how to build AI agents using Microsoft Copilot Studio. The session covered the fundamentals of creating AI agents, including defining goals and objectives, configuring knowledge sources, and setting up tools and protocols. Dinesh demonstrated the process step-by-step using an invoice analyst agent as an example, explaining how to define instructions, select appropriate LLM models, configure knowledge sources from OneDrive or websites, and set up MCP protocols for various tools like email and Power Automate. Students discussed access issues with the Copilot Studio platform and received guidance on using TeamAcademy student accounts to overcome corporate restrictions. The class also addressed questions about prompt engineering, the difference between large and small language models, and how to effectively structure AI agent instructions. Dinesh emphasized the importance of identifying specific business processes to automate rather than focusing on the technical implementation details, comparing AI implementation to early electricity usage where the focus should be on applications rather than the underlying technology.
Copilot Studio Access Issues
The team discussed access issues to the Copilot Studio for creating AI agents. Team members, including IAlfraij and Angela, reported difficulties accessing the platform through their company accounts. The facilitator instructed Angela to use the TeamAcademy student account and provided guidance on accessing the correct credentials, which were shared in the WhatsApp group description. The facilitator emphasized the importance of being able to access and test the platform independently for the class activities.
AI Class Implementation Planning
Dinesh explained to Mehtaji that while Copilot AI Agents require specific studio licenses that may not be available in corporate environments, Mehtaji can still practice with a student account for creating notebooks and projects. Dinesh outlined two objectives for the class: teaching effective use of ChatGPT and Claude, and continuing with the curriculum on AI agent creation and MCP protocols. He emphasized the importance of identifying which departmental processes can be augmented or automated with AI, using the example of an HR recruitment agent that could analyze resumes stored in OneDrive.
AI Applications in Business Processes
Nithin discussed various applications of AI in business processes, including an AI agent for HR recruitment that evaluates candidate qualifications against job requirements, an invoice validation agent that checks compliance with company policies, and a profitability analysis agent for Al-Fardhan exchange house that analyzes transactions by branch and currency. He emphasized that the key to successful AI implementation is understanding how to apply these tools to specific business processes rather than focusing on the technical details of AI development. Nithin also highlighted the importance of working with metadata rather than raw data and explained that effective AI agents should be designed for specific, focused tasks with clear instructions.
Related Offerings
AI Agents in Project Management
Emad asked about creating AI agents for different roles in project management, and Dinesh advised creating one specific role agent at a time rather than multiple agents to avoid confusion. Dinesh explained that while AI agents can perform tasks like email sending, dashboard creation, and database connectivity, they should be connected to a master agent for coordination if multiple specific agents are created. Isa inquired about the difference between large and small language models, and Dinesh clarified that large language models like GPT-5 handle broad knowledge and processing while small language models are customized for specific tasks using focused knowledge bases.
Specialized AI Agent Development Discussion
The team discussed creating specialized AI agents with specific language models focused on particular business processes or procedures. They explained that prompts are essential for instructing AI agents and should be based on existing roles, responsibilities, or business documentation. Angela asked about assessing and sharing AI agents for certification, and the team clarified that to receive a certificate from Team Academy, students need only submit the AI agent instructions via email, while the Microsoft certification requires paying a fee and following self-paced learning materials on learn.microsoft.com.
AI Agent Development Planning
The team discussed creating AI agents and explored various deployment options, including robots that can be programmed with AI instructions. The instructor provided a 15-minute break for participants to think about specific AI agent objectives they would like to create, with the goal of building an AI agent from scratch in Copilot Studio during the class. The session was scheduled to resume after the break to begin the hands-on agent creation process.
Invoice AI Agent Development
The team discussed creating an AI agent with a focus on invoice analysis. The instructor outlined six main steps for creating an AI agent, including defining goals and objectives, configuring the model (recommending GPT-4 reasoning for analysis tasks), and setting up skills and tools. The specific example used was an "Invoice Analyst Agent" that would extract invoices from a defined folder, calculate total costs by product, answer questions about invoices, and send weekly analysis reports on Sundays at 9am. The instructor emphasized that instructions should be written by humans rather than generated by AI, and explained the concept of guardrails to define an agent's role and boundaries.
AI Agent Configuration in Copilot
The team discussed creating AI agents in Copilot Studio, focusing on configuring knowledge sources and tools. They explained how to upload documents (like PDF invoices) as knowledge and demonstrated connecting email tools through MCP (Model Context Protocol) to automate weekly report sending. The instructor clarified that knowledge refers to the data or information the agent should be aware of, while skills and tools are separate components. The session concluded with a note that the next class will cover detailed explanations of ChatGPT and Claude, as well as AI Agent Framework implementation.
Ready to turn real business processes into practical AI solutions?
👉 Explore our AI Transformation Training Program and learn how to design AI agents, configure knowledge sources, select models, connect tools, and automate business workflows using Microsoft Copilot Studio.
