An AI agent is a software-based assistant that uses AI to automate and execute business processes, either working alongside people or acting on their behalf. Instead of just answering a single question, an agent can follow instructions, connect to your business data, and take actions in your systems. You can think of agents in three broad levels: 1. **Simple retrieval agents** These agents focus on finding and explaining information. They can: - Retrieve information from your company’s knowledge sources - Reason over that content - Summarize it - Answer user questions in natural language Example: an **IT Helpdesk agent** that answers questions like “How do I connect to the corporate network?” by pulling from your IT documentation. 2. **Action and workflow agents** These agents go beyond answers and start doing work for you. They can: - Take actions when asked - Automate repetitive workflows - Replace manual, routine tasks for users Example: a **Budget Manager agent** that reviews outstanding open purchase orders (POs) and helps with financial planning. 3. **Advanced, autonomous agents** These agents operate more independently. They can: - React automatically to triggers (like a new email or a record change) - Dynamically plan what to do next - Orchestrate other agents - Learn from context and escalate when needed Example: a **Customer Support agent** that notices when a CRM record is modified, identifies new support issues, and routes them to other agents. Across your organization, you can design agents for many functions: - **Customer support** – handle product questions, refunds, and replacements. - **Finance** – review open POs and support planning. - **HR** – support new hires when an onboarding record is created. - **Operations and contracts** – run scheduled checks or manage contract workflows. In short, an AI agent helps your team move from manual, repetitive work to more automated, guided processes, while still keeping people in control.

AI agents can be set up to act in two main ways: by **following explicit instructions** and by **responding to triggers** in your business systems. 1. **Instruction-based behavior** You define what the agent should do in plain language. For example, for a customer support agent you might specify: - “The goal is to help customers with product questions, refunds, and replacements.” - “If the customer is struggling with the product and asks for a refund, proactively provide relevant product guides and information first.” - “Search for information on the customer in internal systems.” - “For returns and replacements, perform a check on the customer and order details before proceeding.” The agent then follows these instructions consistently, so your process is applied the same way every time. 2. **Trigger-based behavior** Agents can also be configured to react automatically when something happens in your systems. Typical triggers include: - **When a record is modified in CRM** – e.g., a support case is updated. - **When a new email arrives in Office 365 Outlook** – e.g., a refund request. - **When an item is created in SharePoint** – e.g., a new document or request. - **When a new hire starts in HR systems** – e.g., kick off onboarding tasks. - **On a recurrence schedule** – e.g., a contract agent that runs checks daily or weekly. 3. **What actions can agents take?** Once triggered, agents can work with your data and applications. Typical actions include: - **CRM actions** – get a CRM record, update a CRM record, triage support issues. - **Order and file actions** – get order history, create files, log outcomes. - **Collaboration actions** – send a message in Teams, respond in Business Chat. - **Operations actions** – route tasks to other agents, escalate approvals (for example, a returns agent might say: “I have received a refund request that is over 200 and requires your approval.”). You control when agents act (via triggers) and how they behave (via instructions), so they fit your existing processes rather than forcing you to change everything at once.

AI agents connect to your environment through data connectors and a secure platform that’s designed to work with Microsoft 365 and your business systems. 1. **Connecting to your data and apps** Agents can use a range of data sources and connectors, including: - **Microsoft Dataverse** for core business data. - **Microsoft Graph** for productivity data from Microsoft 365 (emails, calendars, files, etc.). - **SharePoint** for documents and list items (for example, when an item is created). - **CRM systems** to read and update records. - **External systems** via data connectors and APIs. With these connections, agents can: - Retrieve knowledge to answer questions. - Take actions like updating records, creating files, or sending messages. - Orchestrate workflows across multiple systems. 2. **Using Microsoft Copilot Studio** Microsoft Copilot Studio provides an end-to-end web platform where you can: - Design your agents’ behavior using a **visual canvas** and **natural language**. - Start from **pre-built templates** (for example, a returns agent or operations agent). - Configure orchestration between multiple agents. - Manage the full lifecycle of your agents. The platform includes: - **Enterprise data protection** - **Granular governance controls** - **No infrastructure management** required - **Streamlined lifecycle management** for deploying and updating agents 3. **Where agents show up for end users** End users can access agents in the tools they already use: - Extend **Microsoft 365 Copilot in Business Chat** with your own agents. - Interact with agents in **Teams** and other channels. - Connect to ISV solutions through the Copilot Studio platform. 4. **Working with Azure AI** Developers can integrate agents with **Azure AI** to add more advanced capabilities while still managing everything through Copilot Studio. In practice, this means you can reimagine existing processes—like customer support, HR onboarding, or contract management—by connecting agents to your current systems rather than rebuilding them from scratch.