OTel needs execution proof

OpenTelemetry is becoming the default way to watch artificial intelligence agents run, but engineers are now drawing a line between seeing a trace and proving an action actually happened. (opentelemetry.io, techcommunity.microsoft.com, dev.to) OpenTelemetry, or OTel, is a vendor-neutral framework for collecting traces, metrics, and logs from software. Its own documentation says it is supported by more than 90 observability vendors. (opentelemetry.io) In OTel’s model, a trace shows the path of a request through an application, and a span represents one operation inside that path. The project’s tracing docs describe traces as the “big picture” of what happens when a request moves across a system. (opentelemetry.io, opentelemetry-python.readthedocs.io) That works well for debugging latency, failures, and tool calls. It does not automatically create a durable, tamper-evident record that can later answer which exact inputs, runtime, and outputs produced a decision. (techcommunity.microsoft.com, dev.to) Arrotu made that argument explicitly in March, writing that most teams rely on “logs, traces, dashboards, and database entries” that help with monitoring but are “not the same as durable, independently verifiable execution evidence.” The post said the gap grows when agents are used in lending, compliance, fraud review, and other workflows with lasting consequences. (dev.to) The timing matters because OTel adoption is spreading into agent tooling, not shrinking. On April 9, Microsoft published a walkthrough showing Azure App Service instrumented with OpenTelemetry GenAI semantic conventions and surfaced in Application Insights’ new Agents view. (techcommunity.microsoft.com) Microsoft said that Agents view shows agent runs, token usage, tool calls, and latency, and that it can light up for telemetry coming from Microsoft Agent Framework, Azure AI Foundry, Copilot Studio, or third-party frameworks that follow the conventions. Azure’s Application Insights docs separately describe OTel support as a standard way to collect telemetry data across platforms. (techcommunity.microsoft.com, learn.microsoft.com) The split is now clearer: observability tells operators what the system reported while it was running, while execution evidence aims to preserve what ran in a form that can be checked later. Arrotu argued for an added execution-evidence layer on top of existing telemetry rather than a replacement for OTel. (dev.to) That leaves OTel in a familiar position for fast-moving infrastructure standards: widely deployed, useful in production, and still incomplete for audit-grade agent records. The next fight is not whether teams want traces, but whether traces alone will satisfy auditors, investigators, and customers months after an agent acts. (opentelemetry.io, dev.to, techcommunity.microsoft.com)