Improving operational efficiency with operations agents in Real-Time Intelligence
We are witnessing the rise of agentic systems: AI-powered agents that perceive, reason, act, and learn in continuous feedback loops. Fabric Real-Time Intelligence is the platform layer that powers them. It connects and makes sense of signals across all different sources (time and space, physical and digital) and can scale to the size and speed of it.
There are many updates from Real-Time Intelligence at Ignite – you can read all about them in Yitzhak Kesselman’s blog post From Data Platform to Intelligence Platform: Introducing Microsoft Fabric IQ.
With the introduction of operations agents, users can now create autonomous agents that monitor data, infer goals, and recommend actions. In future, they’ll also reason over and investigate issues, learning from the issues and approvals they receive. These agents dynamically construct plans based on business objectives, data sources, and available actions, keeping a human in the loop while enabling automation when desired. This helps customers drive operational efficiency, reduce risk, and enhance decision-making.
Getting started with operations agents
Users give the agent access to specific Eventhouse sources, define business goals and instructions, and specify actions integrated through Power Automate. The goals you give it should explain the context for the business process and what the agent should be optimizing for, and the instructions need to help it understand the data, conditions it should look for, and actions to take. As with any LLM tools, giving precise instructions such as specific fields to monitor or specific actions to take help guide it.
The agent then builds a plan to achieve those goals. It makes a ‘playbook’ with the entities and properties it needs to keep track of. It sets up monitoring rules, always grounded with data from the Eventhouse, and then watches for events that match those rules behind the scenes.
When those conditions are met, the agent wakes up and starts to reason over the data. It looks at the actions it’s been configured with and the most appropriate action, and makes recommendations back to the user, along with context about what caused the alert to fire. Any parameters needed to take those actions are dynamically entered based on the data that met the rule’s condition. The recommendation is then presented to users through Teams, which keeps a human in the loop to validate the agent’s actions before it actually takes them.
Next steps
If you want to use Operations agent, you’ll need to enable the Copilot/AI and operations agent tenant settings. You also need a workspace backed by a Fabric capacity; Trial capacities are not supported.
To find out more, refer to the documentation on Operations Agent. We’d love to hear from you about the scenarios you want to use the agent in, and any other feature requests or feedback you have.
Comment on this blog or post in the Fabric Community Forums.
