Operations | Monitoring | ITSM | DevOps | Cloud

AI agent observability means tracing tool calls, reasoning steps, and handoffs, not just tokens and latency. Here's what to instrument and why. Sejal Pandey works on content and growth at Last9, writing about observability, reliability, and SRE practices.

SRE automation covers four different jobs: runbook automation, self-healing infrastructure, drift detection, and resilience testing. Sejal Pandey works on content and growth at Last9, writing about observability, reliability, and SRE practices.

Collecting high-quality user feedback on agents, like from thumbs-up or thumbs-down buttons, is an important part of agent development. User feedback is needed for everything from basic gut checks on whether your agents are behaving well to planning and creating robust eval sets. It’s a critical part of Datadog’s Agent Observability, which provides explicit end-user feedback features for collecting and analyzing it.

Observability investigations rarely follow a straight line. A latency question might cause an AI agent to start with a metric, pivot into traces, compare a deployment window, and finish by reducing thousands of logs to a few patterns. Each individual query is easy, but propagating context throughout an entire investigation can be tricky and expensive.

A user opens a ticket: "The Internet is slow." IT checks the network. Bandwidth looks fine, no outages, no alerts firing. They check with the ISP, but nothing on their end either. Everything upstream checks out, and yet the user is still stuck watching a spinning wheel. What rarely gets checked is the one component sitting closest to the problem: the machine's own network interface.

An LLM inference provider hosts open-weight models like Llama, DeepSeek, and Qwen behind a pay-per-token API, handling GPUs, scaling, and serving for you. The same Llama 3.3 70B model ranges from $0.10 to $1.04 per million input tokens depending on who serves it, so provider choice is a pricing decision. Top picks as of September 2026: Groq and Cerebras for speed, DeepInfra for price, Together and Fireworks for breadth, Baseten for custom models.

A request comes in, enqueues a job, and returns. Twenty seconds later the job runs, and it’s slow. You have a trace of the request and a trace of the job, and nothing joining them. Time Detective has always helped you reconstruct what happened. Now we join it up for you, across applications, services, background jobs and infrastructure, even when they’re built in different languages.

If your app pins openai in requirements or a lockfile, read this before you bump the version, not after something breaks.