OpenTelemetry becoming default
OpenTelemetry is hardening into the de‑facto observability layer for cloud-native apps rather than an optional extra. Microsoft added tooling to monitor AI agents on App Service using OpenTelemetry and launched an Application Insights 'Agents View', tying agent runtimes into standard telemetry pipelines. That development pushes telemetry discipline — traces, metrics and logs — into the center of deploying autonomous or semi‑autonomous features. (cncf.io) (techcommunity.microsoft.com)
Most software teams used to treat observability like a car dashboard you bolt on after the engine is built. OpenTelemetry changed that by giving developers one open set of tools to emit traces, metrics, and logs from cloud-native apps instead of wiring each vendor’s format separately. (opentelemetry.io) A trace is a step-by-step receipt for one request moving through a system. If a checkout button triggers five services and one database call, a trace shows which hop slowed down or failed, in order. (opentelemetry.io) (cncf.io) OpenTelemetry grew out of two earlier projects, OpenTracing and OpenCensus, and the Cloud Native Computing Foundation accepted it in May 2019. By August 2021, it had moved to incubating status inside the foundation, which is where widely used production infrastructure projects usually harden. (opentelemetry.io) (cncf.io) The pitch was simple: instrument your code once, then send the data anywhere. The official OpenTelemetry site says the same application data can be exported to Jaeger, Prometheus, commercial vendors, or an in-house backend without changing the app code again. (opentelemetry.io) That mattered for ordinary web apps, but artificial intelligence agents create a messier problem. One agent run can include model calls, tool calls, retries, handoffs to another agent, and token usage charges, all inside one user request. (techcommunity.microsoft.com) Microsoft is now pushing those agent runs into the same telemetry plumbing. In an April 9, 2026 Azure blog post, the company showed how Azure App Service apps can instrument agents with OpenTelemetry generative artificial intelligence semantic conventions and send that data into Application Insights. (techcommunity.microsoft.com) Semantic conventions are shared labels for telemetry, like agreeing that every shipping box should put the address in the same corner. Microsoft’s agent monitoring uses OpenTelemetry generative artificial intelligence semantics so token counts, tool calls, and agent steps arrive in a format dashboards can understand consistently. (learn.microsoft.com) (techcommunity.microsoft.com) Application Insights now has an Agents view in preview that pulls those runs into one screen. Microsoft says that view can monitor agents from Microsoft Foundry, Copilot Studio, and third-party agents, with telemetry for performance, diagnostics, token usage, and cost analysis. (learn.microsoft.com) This is the shift: the agent is no longer a black box sitting outside the normal production stack. Microsoft’s own setup guides now tell developers to connect Application Insights, instrument agent frameworks, and view traces for agent behavior the same way they would inspect a payment API or search service. (learn.microsoft.com 1) (learn.microsoft.com 2) OpenTelemetry is also expanding beyond the old three pillars of traces, metrics, and logs. The project’s March 26, 2026 blog announced a public alpha for profiles, which are continuous snapshots of where code spends time, pushing the standard deeper into performance work as well. (opentelemetry.io) Put together, that means OpenTelemetry is becoming less like a nice-to-have library and more like the default wiring behind modern software. When a company ships an agent that books travel, calls tools, or hands work to another model, the expectation is increasingly that every step should already show up in standard telemetry. (techcommunity.microsoft.com) (opentelemetry.io)
