Azure Copilot Observability Agent GA: Complete Guide to Features, Autonomous Operations, Pricing, and Limitations

On June 23, 2026, Microsoft announced the general availability of the Azure Copilot Observability Agent, alongside a public preview of autonomous operations. Built on Azure Monitor, the agent is designed to help engineering, SRE, DevOps, and operations teams move from telemetry and alert noise to investigated issues, explainable reasoning, and recommended next steps that can reduce Time-To-Mitigate (TTM) . Here is a complete breakdown of everything announced, including features, pricing, constraints, and how the agent diagnoses outages.

Key Features and Capabilities

The Observability Agent acts as an AI-powered assistant built into Azure Monitor, grounded in real telemetry data . Its core capabilities include:

• AI-driven troubleshooting layer: Correlates logs, metrics, traces, topology, resource health, and operational context across cloud environments .
• Natural-language chat: Allows engineers to ask open-ended questions and conduct guided investigations against telemetry data without hand-writing Kusto Query Language (KQL) queries .
• Deep investigations: The agent frames hypotheses, gathers evidence, compares signals by time/scope/type, rules out weak explanations, and presents a full reasoning trail .
• Reasoning transparency: The agent displays what it found, what it ruled out, and why, supporting human judgment rather than bypassing it .
• Broad signal correlation: Works across Azure-monitored applications, AKS, VMs, Foundry telemetry, infrastructure, and platform signals .
• Azure RBAC integration: Interactive chat and investigations use the signed-in user's identity and role-based access control .
• No training on user data: Prompts and responses are not used to train foundation models .

Autonomous Operations (Public Preview)

Alongside GA, Microsoft introduced autonomous operations in public preview. These capabilities run in the background to reduce manual triage work:

• Continuous alert correlation: The agent continuously analyzes alerts as they fire, groups related alerts into a single Azure Monitor issue, and attaches a natural-language explanation of why those alerts belong together .
• Issue creation from prominent alerts: It can create issues from individual significant alerts based on custom instructions (e.g., "always create an issue for severity 2 alerts") .
• Automatic deep investigations: The agent can run deep investigations on issues it has created .

Crucially, humans remain in control. The agent does not restart resources, change configuration, or resolve issues on its own . All mitigation decisions stay with the engineer . Autonomous operations are subject to the Supplemental Terms of Use for Microsoft Azure Previews .

How the Agent Diagnoses Cloud Outages

The agent is informed by Microsoft's years of experience running Azure. It is designed to diagnose what is going wrong during outages without the stress, fatigue, or tunnel vision that can affect human operators .

It connects logs, metrics, traces, alerts, dependencies, resource graph, resource health, activity logs, Foundry telemetry, and changes to trace a problem back to its likely cause . Kubernetes co-creator Brendan Burns described the agent as adding a hypothesis-formation layer on top of deterministic automation: the agent can form a theory about what went wrong, test it against the data, and continue working to find a solution .

Limitations and Constraints

Several constraints apply to the current release:

• Autonomous operations are in public preview only — they are not covered by GA support terms and are subject to preview legal terms .
• No autonomous action: The agent investigates and recommends but does not restart resources, change configuration, or resolve issues .
• Custom instructions limit: Custom instructions are limited to 8,192 characters per Observability Agent resource .
• Deep investigation cap: A single deep investigation operation is capped at 500 Azure Agent Credits (AAC) .
• No dedicated filter for agent-created issues: In the issues list view, only the [<AppName>] title prefix distinguishes agent-created issues .
• Region and scope constraints: Autonomous operations run against a dedicated Microsoft.Monitor/observabilityAgents resource with a scoped managed identity .

Pricing and Billing

• Consumption-based pricing model: You pay only for the AI work the agent performs, measured in Azure Agent Credits (AAC) .
• Three billing patterns: Chat (usually lowest cost), deep investigations (higher cost, capped at 500 AAC per operation), and autonomous operations (alert correlation is not billed during preview; automatic deep investigations are billed) .
• AAC list prices: Available on the Azure Monitor pricing page .
• Billing start date: July 1, 2026 .
• Alert correlation: The continuous background grouping of alerts is not billed during public preview .

Note on KPMG

Available evidence covers KPMG's rollout of Agent 365 (Microsoft's broader agent governance platform), not the Observability Agent specifically. KPMG deployed Agent 365 to 276,000 people as part of an enterprise-wide agent governance initiative . No specific results or metrics for KPMG's use of the Azure Copilot Observability Agent itself were found in the provided sources. Microsoft's blog posts and community articles highlight the GA announcement but do not cite KPMG as an early adopter of this particular agent.