Infrastructure & Automation Weekly

What Modern DevOps Looks Like in 2026

The DevOps role has fractured. Platform engineers manage internal developer platforms. SREs handle reliability and observability. Infrastructure architects design multi-cloud deployments. Security engineers harden the stack. This newsletter is for the people who actually run production infrastructure, manage Kubernetes clusters, optimize cloud costs, and build the automation that keeps engineering teams moving.

Kubernetes Has Won. Now What?

Kubernetes adoption crossed a tipping point in 2025-2026. The question is no longer whether to use it, but how to use it without burning engineering cycles on platform maintenance. K8s has become the default for container orchestration, yet operational overhead stays high. Clusters are complex, the tooling ecosystem is overwhelming, and cost management is still unsolved for most orgs.

The DevOps engineers doing well here are not running the most sophisticated clusters. They have made deliberate trade-offs: picked specific tools and stuck with them (Helm over Kustomize, ArgoCD over Flux), limited third-party controllers, and baked cost optimization into the initial cluster design instead of bolting it on later. They know their blast radius, can explain their DR plan in thirty seconds, and had observability wired in from day one.

Infrastructure-as-Code: Terraform, Pulumi, and the CDK Wars

The IaC landscape has fragmented into three camps. Terraform owns market share and mindshare, but it is showing its age: HCL syntax limitations, state management gotchas, upgrade pain for large codebases. Pulumi lets you use real programming languages, but trades one complexity problem for another. AWS CDK and Constructs are maturing, but they lock you into one cloud provider.

The right pick depends on your team's velocity, your multi-cloud strategy (or lack of one), and how much patience you have for managing state. This newsletter covers real trade-offs, not marketing. You will read about teams that regretted Pulumi, teams that outgrew Terraform, and teams that committed to CDK with zero regrets.

Observability Without Bankrupting the Company

Datadog pricing in 2026 is a masterclass in product-led growth with a side of taxation. Most teams now pay $2-5K/month for observability that used to cost a fifth as much. New Relic is chasing the mid-market. Splunk is enterprise-only. And everyone is asking the same questions: self-host? Sample more aggressively? Go all-in on open source?

The answer depends on your scale and your team's bandwidth. Self-hosting Prometheus, Loki, and Jaeger works great at 10-50 engineers and becomes a drag at 100+. The cost is not just licensing -- it includes on-call burden, upgrade cycles, and the expertise to keep things running. This newsletter breaks down the real cost of each approach with actual numbers from production.

Cloud Cost Optimization: Where Most Teams Leave Money on the Table

Every engineering org over-commits on cloud resources. Reserved instances, spot instances, committed use discounts, and the perpetual mystery of what you are actually being billed for. The mechanics are complicated by design. FinOps has emerged as a discipline, but most FinOps problems are cultural and organizational, not technical.

The engineers who have cracked this are not micro-optimizing reserved instances. They architected for cost from the start: right-sized resource requests, aggressive scheduling, automated cleanup of orphaned resources. Product teams understand the cost of their choices, not just the performance. And tooling tracks cost attribution at the service level, not just the account level.