Operations | Monitoring | ITSM | DevOps | Cloud

You rely on NetFlow to give you the visibility needed to trace bandwidth consumption, identify suspicious traffic patterns, and plan for future capacity requirements. However, monitoring flow data has grown increasingly complex over the past few years. As enterprise environments expand into hybrid architectures and user traffic volumes multiply, capturing and processing this data creates operational challenges.

It's 2:14 AM. CPU usage is normal. Memory looks stable. No pods are in CrashLoopBackOff. Every dashboard is green. And yet API latency has doubled, checkout requests are timing out, and your on-call phone won't stop buzzing. This is the defining trait of abnormal Kubernetes workload behavior: it rarely announces itself through the metrics you already watch. Kubernetes is exceptionally good at reporting whether a pod is running. It is far less good at telling you whether a pod is doing its job correctly.

When you're the on-call engineer and something breaks, you can quickly find yourself deep in a series of tools you don't regularly use—switching tabs, copying query results, and manually stitching together a picture of what's happening and why. People are increasingly turning to AI to get around this, but the results can be a mixed bag.

The world is especially hard right now. The future of the software engineering profession looks more uncertain than ever. Execs are under heavy pressure to turn AI into magic results, and teams are fighting product competition and AI-induced burnout on one side, melting mental models and hellish oncall on the other side. Observability was supposed to be a solved problem by now.

The last few years have seen AI conversations dominated by the need for investment in hyperscale infrastructure as firms race to build ever larger training models. But as those conversations evolve, the emphasis is shifting to the next phase of AI adoption, focusing on the scaling of use cases and real-world value. In line with this shift, organisations are looking beyond where AI is trained to the specifics of where it is actually used.

As Staff Engineers and Principal Architects, most of us have spent years thinking about long-term system health. We are considerate of the company’s business objectives and strategy, accumulation of technical debt, and operational risk. For us it is not about whether code works today, but whether the engineer who inherits it in three years will be able to understand what it was trying to do and why. That's what makes a codebase maintainable rather than just functional.

AI cost observability is the practice of measuring, attributing, and analyzing AI workload costs at the request, model, and workflow level in real time. It connects cloud infrastructure spend, inference and token costs, and business attribution (cost per feature, team, customer, or product) so engineering, finance, and product teams can see where AI spend goes and whether it creates value.

Two-dimensional group-by improvements in Explorer.

AI cost observability is the practice of measuring, attributing, and analyzing AI workload costs at the request, model, and workflow level in real time. It connects cloud infrastructure spend, inference and token costs, and business attribution (cost per feature, team, customer, or product) so engineering, finance, and product teams can see where AI spend goes and whether it creates value. On July 14, IBM had its worst trading day since 1987.

Automation has a branding problem. For years, it has been associated with cost reduction and workforce replacement. But operators tell a different story. Across eleven interviews, the consistent theme was relief. Relief from manual ticket creation. Relief from repetitive triage. Relief from workflows that once required three days and now take five minutes. These are not stories about eliminating people. They are stories about protecting them. Operators spoke with clear ownership over their environments.