At CollabDays Finland 2025, Microsoft MVP Renato Romão delivered a session that was both deeply technical and highly practical. With nearly five years of experience working with Copilot Studio, Renato brought clarity to a rapidly evolving space: AI-powered automation inside the Power Platform.
His session was live demonstration of how to build, deploy, and scale autonomous agents using Copilot Studio. And best of all, he shared working examples via GitHub, allowing attendees to download and test real solutions.
🤖 What Is Copilot Studio?
Renato began by introducing Copilot Studio as Microsoft’s evolving platform for building intelligent agents. Originally launched as a virtual agent tool, it has now matured into a full-featured environment for creating autonomous, voice, and assistant agents—all without writing code.
He explained the difference between:
- Copilot: AI assistants that help with productivity (e.g., answering questions)
- Agents: Custom-built bots that automate business processes (e.g., reading invoices, responding to emails)
Copilot Studio allows users to define triggers, tools, knowledge bases, and agent flows—all orchestrated to deliver intelligent automation.
🛠️ Live Demo: Two Real Agents
Renato showcased two agents he built specifically for the event:
1. Feedback Agent
- Triggered by incoming emails
- Analyzes session feedback
- Responds automatically in a defined tone and language
- Handles multilingual input and avoids filtering negative feedback
2. Invoice Agent
- Processes PDF invoices sent via email
- Extracts key data using AI
- Stores or responds with relevant information
- Can be reused across clients by exporting the solution
These agents were built using Copilot Studio’s no-code interface, and connected to Power Automate, Outlook, and other Microsoft 365 services.
💡 Best Practices for Agent Design
Renato emphasized several key principles when building agents:
- Context matters: AI needs clear instructions to understand the scenario
- Tone and language: Define whether responses should be formal/informal and in which language
- Step-by-step logic: Break down workflows into clear, actionable steps
- Security first: Use Microsoft’s secure connectors instead of exposing enterprise data to external AI platforms
He also highlighted the importance of Copilot credits—a licensing model that determines how much automation you can run per tenant. For example, basic actions may cost 1 credit, while advanced flows could consume 13 or more.
📦 Deployment & Scalability
One of the session’s most valuable insights was how easily agents can be exported and reused across environments. Renato showed how to:
- Build agents in a developer environment
- Package them as solutions
- Deploy them to different clients or departments
- Connect them to enterprise data securely
This makes Copilot Studio a powerful tool not just for experimentation, but for enterprise-grade automation.
🔊 Voice Agents & Advanced Scenarios
Renato also introduced Voice Agents, a newer feature in Copilot Studio that allows users to interact with bots via phone calls. These agents can route calls, respond to voice prompts, and integrate with backend systems.
He discussed advanced use cases like:
- Customer support triage
- Multi-agent orchestration
- Integration with Azure AI Foundry for custom models
“It’s not just about building bots—it’s about solving problems with AI,” Renato said.
His GitHub repository includes both agents and setup instructions, making it easy for developers and business users alike to get started.