AzureFeeds

Observability for AI and Agent Workloads

AI agents are moving from prototype to production. To help teams ship and operate agents with the same rigor as the rest of their stack, Azure Monitor & Azure Copilot Observability agent brings end-to-end agent observability — grounded in OpenTelemetry so signals are portable across the toolchain.

Agent Observability in Azure Monitor

Agents are now a first-class artifact in Azure Monitor. With new views showing agent fleet, automated evaluations, cost breakdown, trace tree, and human-in-loop evals, you have the tools you need to gain observability in your agent. Microsoft Foundry is where you build your agents and set up evals, and Azure Monitor is your full-stack observability solution across all layers and components of your distributed service. Now, it’s easier to get started with a streamlined Microsoft OpenTelemetry Distro that powers all observability + governance surfaces including Foundry, Azure Monitor, and Agent 365.

Learn more: aka.ms/agent-obs-blog  |  Trace agents with the Agent Framework

Azure Copilot Observability agent – what’s new

The Observability agent, part of Azure Monitor, enables engineers to investigate issues and explore system behavior using natural language over telemetry data. At Build 2026, new updates expand its capabilities across both chat and investigation workflows. These enhancements include broader investigation entry points (such as AKS and Application Insights), deeper cross-resource analysis, and integration with Microsoft Foundry AI agents. Together, they provide end-to-end visibility into AI-driven systems, help teams move faster from detection to root cause, and enable sharing of investigation results for collaboration and follow-up.

Learn more: Azure Copilot observability agent (preview) – Azure Monitor | Microsoft Learn  | 

https://aka.ms/ObsAgentBlogBuild26

ROI of Agents in Foundry (Public Preview)

A new ROI view in Microsoft Foundry quantifies the business value of deployed agents — correlating cost, usage, and outcomes — so teams can see which agents are paying off and where to invest next.

Smarter, simpler monitoring with Azure Monitor

As cloud environments grow, monitoring gets harder. There is a need for reduced alert noise, less manual tuning, tighter security, and monitoring that is accessible so you can catch real issues faster and spend less time managing the tool.

Resource-scoped querying of Azure Monitor Workspace metrics — Generally Available

Azure Monitor workspaces now offer users the ability to scope their PromQL queries to one or more Azure resources (e.g., Virtual Machine, AKS, Application Insights, etc.) without requiring the user to have direct access to the AMW(s) where metrics are stored, to streamline the user experience and offer parity with how resource-scoped queries work on Log Analytics Workspaces today.

Learn more: Resource-scoped queries for Azure Monitor workspace – Azure Monitor | Microsoft Learn

Dynamic thresholds for log search alerts — Generally Available

Log search alerts now support dynamic thresholds at GA, using machine learning to learn each rule’s normal behavior from historical query results and automatically account for hourly, daily, and weekly seasonality. Thresholds are calculated per dimension combination, so multi-dimensional scenarios like AKS pod-restart spikes or resource-inventory drift get tailored baselines out of the box — with no manual tuning and no extra charge beyond the standard log search alert rate.

Learn more: Alert rules with dynamic thresholds overview

Simple log alerts in Azure Monitor — Generally Available

Simple log alerts are a type of log search alert that evaluates each row individually instead of aggregating over a time window, delivering low-latency detection for scenarios like failed automation jobs or critical Windows events. They also support Basic Logs, so customers can keep the cost savings of Basic-plan telemetry — including Application Insights traces — without giving up the ability to alert on it. Flexible trigger recurrence lets teams tune sensitivity and reduce noise without sacrificing responsiveness.

Learn more: Create a simple log search alert in Azure Monitor

Expanded OpenTelemetry Support

Modern environments span many clouds, languages, and platforms, and customers have increasingly standardized on OpenTelemetry (Otel) to instrument them consistently. The challenge is turning all that telemetry into real insight. Azure Monitor brings OpenTelemetry metrics, logs, and traces into one place where teams can troubleshoot quickly, visualize what’s happening, and act on it. From VMs and servers to applications and AI coding agents, get faster triage, troubleshooting, and ready-made dashboards all from one centralized experience in Azure.

OpenTelemetry App Troubleshooting via OTLP Ingestion — Generally Available

Azure Monitor now offers flexible, enterprise-ready OpenTelemetry ingestion and data storage to power application performance monitoring experiences. Use standard OpenTelemetry instrumentation and OTLP export to send metrics, logs, and traces to Azure Monitor data collection endpoints. Then monitor, triage, and troubleshoot application and platform performance using Application Insights and pre-built Grafana dashboards entirely based on OpenTelemetry data.

Learn more: https://aka.ms/AzureMonitorOTLPDirectGAblog |  Ingest OpenTelemetry data into Azure Monitor.

Monitor AI coding agents with OpenTelemetry (GA)

With Azure Monitor’s OpenTelemetry support, you can collect OpenTelemetry Protocol (OTLP) signals from AI coding agents such as GitHub Copilot and Claude Code, and route them into Azure Monitor for end-to-end visibility. Ingested OTLP data is stored with OpenTelemetry semantics for logs and traces. Application Insights provides curated agent views for troubleshooting, detailed trace visualizations, end-to-end transaction views, and dedicated Grafana dashboards for coding agent monitoring.

Once OpenTelemetry metrics are ingested in Azure Monitor, they can be used to create SLIs.

Learn more: https://aka.ms/AzureMonitorOTLPCodingAgentsblog |  https://learn.microsoft.com/en-us/azure/azure-monitor/app/agents-view

OpenTelemetry Metrics, Visualizations, and Enhanced Monitoring for Azure VMs and Arc Servers — Generally Available

Azure Monitor now supports OpenTelemetry (OTel) metrics and visualizations for Azure Virtual Machines and Arc-enabled Servers, delivering an enhanced, unified monitoring experience. This release brings together key monitoring capabilities including recommended alerts, out-of-the-box Grafana dashboards, and at-scale configuration into a single experience. Customers can more easily monitor Guest OS health, accelerate troubleshooting, and optimize both performance and monitoring costs across their environments.

Learn more: https://aka.ms/vmiv2docs | Collect and customize OpenTelemetry metrics for Azure virtual machines – Azure Monitor

One Data Platform for Any Source and Any Destination, built on OpenTelemetry

OpenTelemetry brings consistency to telemetry collection, but production systems need control over how that data is processed and delivered. Learn how Azure Monitor enables centralized governance, multi-stage transformations, unprecedented scale (billions in EPS), and flexible routing for all your telemetry to turn OTel signals into actionable observability.

Learn more: https://aka.ms/datacollectionbuild2026

SLI and SLO — Generally Available

Azure Monitor now supports service-level indicators and objectives. These special metrics can now measure availability and latency from an end user’s critical journey. Define an SLI across the resources (Service Groups) that make up a service, set an SLO target on your Azure OTel and Prometheus metrics, and track error budget and burn rate from a single view. This lets engineering and SRE teams align day-to-day work to the customer experience, not just per-resource health.

Learn more: Create service level indicators in Azure Monitor.

Get started

We’re continuing Azure Monitor’s investment in efficient, end-to-end observability for developers, SREs, and IT pros. To learn more, connect with our experts at the Build Lightning Session- Broken, costly? Debug and operate AI agents with Azure Monitor on June 3^rd at 9:50 AM PT.