Microsoft Fabric Blog explains how operations agents leverage Copilot and autonomous AI reasoning to monitor and act on data, with guidance on usage metering, billing, and technical metrics for architects, engineers, and admins.

Understanding Operations Agent Capacity Usage and Billing in Microsoft Fabric

Microsoft Fabric’s operations agents, announced at Ignite, introduce autonomous capabilities for data monitoring, goal inference, and actionable recommendations. As Preview continues, billing for these agents will be enabled, aligned with Fabric Capacity Unit (CU) consumption, and reported in the Capacity Metrics App.

Key Operations Agent Activities

Copilot in Fabric Usage: Charges accrue when Large Language Model (LLM) features are used for agent configuration or direct interaction.
Compute: Covers background querying and monitoring, with the agent evaluating rules and conditions. This compute is tracked in the metrics app, with potential additional costs from data sources.
Autonomous Reasoning: When monitored criteria are met, the agent leverages LLMs to summarize data, recommend actions, and communicate with users.
Storage: Fabric retains monitored data and agent configurations for 30 days, incurring storage costs based on OneLake rates.

Capacity Metrics Tracking

Configuration, query, rule tracking, and data retention all generate usage, tracked in the Capacity Metrics App:

Long-lived queries against Eventhouse result in both compute and storage costs.
Autonomous reasoning events charge by token usage, reflecting deep AI processing for recommendations.

Flow Diagram (Overview)

Configure agent: Copilot creates playbooks, Eventhouse queries identify data fields; storage tracks configuration.
Activate agent: Agents run background queries, apply rules; compute and storage meters track usage.
Trigger recommendations: LLM summarizes matching data; autonomous reasoning CUs calculated; user can approve Power Automate flows (consult latest Power Automate pricing).

Usage Categories and Rates

All operations agent activity in Fabric is considered Background usage (including direct interaction). Copilot and AI operations are billed under background, detailed in the Copilot Fabric Consumption docs.

Azure Metric Name	Fabric Operation Name	CU Rate
Operations Agents Compute Capacity Usage CU	Operations agent compute	0.46 CUs per vCore hour
Copilot and AI Capacity Usage CU	Copilot in Fabric	100 CUs/1000 input tokens; 400 CUs/1000 output
Operations Agents Autonomous Reasoning Capacity Usage	Autonomous reasoning	400 CUs/1000 input; 1600 CUs/1000 output
n/a	OneLake Storage	per GB per hour (link)

Timeline

Usage reporting visible in Capacity Metrics App from December.
Billing for Copilot in Fabric starts December; other features billed after January 8, 2026.