Operations | Monitoring | ITSM | DevOps | Cloud

Are you writing agentic applications, but aren’t sure what the agents are doing? Finding out too late that you've blown the budget with super expensive models? Not sure where the agents are failing, and feeling a loss of control? Could they do better? Observability is the visibility you need to get the job done. Sending telemetry to Honeycomb explains what your agents are actually doing.

As organizations continue to heavily invest in AI and build more agentic workflows, their telemetry data volumes can surge quickly, and the associated costs can become unpredictable. To regain control of their data, many AI-forward teams are turning to high-throughput, low-latency pipelines to collect and route data to tools such as OpenTelemetry (OTel) and ClickHouse. But these self-hosted solutions come with drawbacks.

Single Step Instrumentation (SSI) simplifies Datadog Application Performance Monitoring (APM) by automatically discovering and instrumenting services across a host. For many teams, SSI is the ideal starting point because it helps them achieve full visibility with minimal setup. However, as environments grow, teams often want more control over which services get traced. Auxiliary workloads such as batch jobs and cron tasks might not require distributed tracing.

N+1 query problems are one of the most common, and quietly damaging, performance issues in production applications. One extra query per record feels harmless in development. At scale, it becomes the reason your response times degrade and your database buckles under load. Today, AppSignal adds N+1 detection to its OpenTelemetry support. When we identify the pattern in a trace, we collapse the repetitive spans directly in the timeline, making the problem immediately visible in the trace itself.

Slow endpoints are difficult to detect because they don’t fail. They simply get slower and slower. Average latency may look fine, but that can be misleading. That’s why we need to look at other values, like p90 and p95, which often reflect what’s really going on. For example, p90 represents the slowest 10% of requests, and p95 represents the slowest 5%. When these values increase, users start experiencing delays.

Agentic workloads thrive with precision tooling. Just like developers, they need the rich context, high cardinality, and fast feedback loops that allow them to ask exploratory open-ended questions of their code. But instrumentation is costly, and from the dawn of software, developers have tried to do the most possible with the least amount of resources.

OpenTelemetry is great at answering one question: “what just broke?” The problem is that most teams need a different answer first: “what is about to break in this release?” That is where trace-based testing comes in, especially for teams running a vendor-neutral OTel stack (Collector + Tempo/Jaeger + Prometheus) and needing deterministic release gates.