Twenty-five Helm flags, template functions, chart features and tools that most daily users never touch, each with the operational problem it solves. Written after two years of writing and managing Helm packages in production.
Normal code has tests; agents need evals. What an eval harness is, the vocabulary that goes with it, how I would build one in Go for an SRE agent working against Kubernetes, why pass^k matters more than pass@k for ops, the tools that exist (and the one OpenAI is shutting down), the ways a harness's numbers lie to you, and the habits that keep a suite useful.
What Helm does with a hook step by step, the three things hooks do not do that everyone assumes they do, a migration Job that survives a rollback, and when an initContainer or an Argo CD sync wave is the better tool. Written after two years of writing and managing Helm packages in production.
A container log line on disk contains no idea which pod wrote it. Everything you filter by later is constructed by the collector at collection time. What the file actually looks like, how collectors find it, where every label comes from, and why that design decides the health of your logging stack.
Why a changed ConfigMap restarts nothing, the Helm checksum annotation done properly, the two mistakes that make it silently do nothing, and when an immutable ConfigMap, Reloader, or a reload sidecar is the better tool. Written after two years of writing and managing Helm packages in production.
I learned enough Rust in a weekend to read, review and debug the CLIs, API clients and daemons that show up in infra work. This is the subset that mattered, the compiler errors you will actually hit, and the parts of the language I deliberately skipped.
Most LLM calls in a backend are structured decisions dressed up as chat. TypeSafe's Jev answers only that kind of question: typed answers with probabilities, in one pass, with no text generation. What it is, how its confidence numbers work, what 'can't hallucinate' does and does not mean, the open-source alternatives that appeared within weeks, and where an SRE would put one.
Which of my Kubernetes-shaped SRE instincts are about the abstraction and which are about the machine underneath, worked out against a hosting company that runs about 12,000 single-tenant VMs across ten cloud providers with no Kubernetes at all.
VictoriaMetrics started as a remote storage backend for Prometheus and grew into a full monitoring stack. What Prometheus deliberately does not do, why two HA Prometheus replicas are not a cluster, how VictoriaMetrics splits Prometheus's jobs into components and keeps long-term data without object storage, how VictoriaLogs compares with Loki and VictoriaTraces with Tempo and Jaeger, what else the project ships, and when it is the right choice.
I read the whole of Regulation (EU) 2024/2847, every article and every annex, and mapped KubeAid and LinuxAid against it for my employer. These are the things that surprised me, the mental model I ended up with, and what I would tell the next engineer who has to do the same reading.