Enabling agents in Microsoft 365 Copilot Chat 

Our ambition is to empower every employee with a Copilot and to transform every business process with agents. Today, we’re continuing to deliver on that ambition with Microsoft 365 Copilot Chat. Copilot Chat enables your entire workforce—from customer service representatives to marketing leads to frontline technicians—to start using Copilot and agents today. Read the full announcement here. 

Agents play a significant role in how organizations are transforming their business process with AI. Soon agents created in Copilot Studio agent builder and Copilot Studio will be available right from within Copilot Chat. Let’s review how you can pay for agents with Copilot Studio.

Getting started with agent consumption 

Agents play a significant role in how organizations are transforming their business process with AI. Soon agents created in Microsoft Copilot Studio agent builder and Copilot Studio will be available right from within Copilot Chat. Let’s review how you can pay for agents with Copilot Studio. 

Usage of agents is measured in “messages” and total cost is calculated based on the sum of messages used by your organization. You can purchase messages though the Copilot Studio meter in Microsoft Azure, a pay-as-you-go option, for $0.01/message, or via pre-paid message packs priced at $200 for 25,000 messages/month.  

The Microsoft Power Platform admin center is used to set up billing and assign message capacity to Copilot Chat and individual agents. Agent message usage varies with an agent’s complexity, frequency, and the use of specific features. These features have been categorized into five groups, including three new ones introduced today. 

Classic and generative answers 

Answers are how agents communicate with users and are core to the prompt and response interaction model. Answers are charged for every agent response, not for user prompts. As makers build agents, they can define conversational topics, create branching logic flows, and apply generative AI to create responses based on knowledge sources. To support this range of capabilities, we provide two options: 

1. Classic answers, used for predefined responses that are manually authored by agent makers. These are static, do not change unless manually updated, and are typically used when precise responses are required. Classic answers cost one message. 

2. Generative answers, used for dynamically generated responses based on knowledge sources and context. These provide more flexible and natural interactions because they build on a conversation’s context and available knowledge. Generative answers cost two messages. 

Note that users licensed with Microsoft 365 Copilot can use classic and generative answers at no cost as part of their per user license. 

Tenant Graph grounding for messages 

Tenant Graph grounding for messages provides high quality grounding for agents with relevant information from Microsoft Graph, including files stored in SharePoint and external data synched via Graph connectors. It does not include personal Graph data such as emails and chats. Tenant Graph grounding for messages costs 30 messages for each knowledge retrieval. 

This paid capability is optional and you can decide whether to turn it ‘on’ or ‘off’ in Copilot Studio. If this capability is enabled, each time an agent responds with a generative answer, it will utilize knowledge from the tenant Graph and tenant Graph grounding for messages will be billed. This will roll out in phases—existing agents that enabled this feature during preview, along with new agents built in Copilot Studio, will be billed starting on February 1, 2025; and new agents built in Copilot Studio agent builder using this feature will be billed starting today. 

Note that users licensed with Microsoft 365 Copilot can use tenant Graph grounding for messages at no cost as part of their per user license. 

Autonomous actions enter paid preview 

Autonomous agents understand your work, act on your behalf, and can automate complex business processes. We’ve seen significant interest in these capabilities since public preview launched in November 2024, and starting February 1, 2025, we are transitioning to a paid preview program. 

To enable autonomous agents, we are introducing the concept of autonomous actions. Autonomous actions are generatively orchestrated triggers, topics, data connectors, and workflows and are visible in the activity map displayed in generative orchestration mode. Autonomous actions cost 25 messages per action. 

Calculating agent cost  

Agent costs are calculated based on the capabilities from the table above. Here are a few example cost calculations: 

• A hypothetical agent might answer questions from customers on a website. Yesterday, it consumed 500 classic answers and 2,000 generative answers. Therefore, it would cost 4,500 messages, equivalent to $45 for that day. 

• A hypothetical agent in Copilot Chat uses data stored in Microsoft Graph to answer employee questions about HR policies. Yesterday, the agent consumed 200 generative answers and 200 tenant Graph grounding for messages. Therefore, it would cost 6,400 messages or $64 for that day.  

• A hypothetical autonomous agent responds to and routes inbound sales orders from customers. Yesterday, it consumed 100 generative answers, 100 tenant Graph grounding for messages, and 800 autonomous actions. Therefore, it would cost 23,200 messages or $232 for that day.  

Of course, every agent will be different, and the number of messages consumed will depend on usage frequency and complexity. You can learn more about these features and billing. 

We recognize how critical it is to provide you with appropriate governance tools to manage agent costs within your organization. Within Microsoft Power Platform admin center, admins can assign agents to environments, monitor agent usage, and allocate message capacity to ensure you stay in control of billing. 

See a quick walkthrough of agent management within Microsoft Power Platform admin center and learn more about agent management.