TLDR DevOps 2026-02-18

What if your observability platform didn't just detect issues, but resolved them automatically?

At New Relic Advance — a free virtual event on February 24 — the team will show how agentic AI plugs into observability workflows to detect, diagnose, and resolve incidents without waiting for a human. They're also demoingfull-stack visibility for AI applications and tools that tie system performance directly to business metrics.

If you're running production systems and want to see where observability is heading, this is worth a calendar block.

Save your free spot: AMER | EMEA | APAC

Amazon EC2 has introduced the ability to create nested environments within virtualized instances, a capability previously limited to bare metal EC2. Available on C8i, M8i, and R8i instances across all commercial regions, this allows customers to run KVM or Hyper-V for use cases such as mobile application emulators and Windows Subsystem for Linux.

The new `terraform-state` provider for Pulumi ESC is designed to directly read outputs from Terraform state files and integrate them as first-class values within ESC environments. This eliminates manual copy-pasting and scripts, automatically treating sensitive Terraform outputs as secrets and supporting remote backends like Terraform Cloud.

Go 1.26 introduces a completely rewritten `go fix` subcommand that automatically modernizes Go code by applying fixes that leverage newer language features, such as the enhanced `new` function, to improve clarity and ensure LLMs are trained on current idioms. This advancement builds upon the Go analysis framework and foreshadows a future "self-service" paradigm for user-defined modernizations.

AWS built an open source Kubernetes controller to integrate Karpenter with Amazon Application Recovery Controller zonal shift, automatically reconfiguring node pools to avoid impaired Availability Zones during failures. The solution improves EKS resiliency, aligns autoscaling with traffic shifts, and can be tested using AWS Fault Injection Service.

Bias toward action means taking the smallest responsible step that generates real user feedback—while building guardrails (feature flags, monitoring, rollbacks, and error budgets) that make being wrong safe and quickly reversible. Fast teams aren't reckless - they invest in infrastructure that makes frequent, low-risk deployments boring, so they can learn quickly without compromising reliability.

RunPod delivers two core building blocks: Pods for hands-on training and experimentation, and serverless inference for production workloads that scale automatically. Provision compute in seconds, deploy models straight from code, and observe everything from a single interface. No long-term commitments, no stitched-together services, just infrastructure designed for developers shipping AI.

roam pre-indexes your codebase into a semantic graph so AI agents can query structure instead of grepping files. Unlike LSPs (editor-bound and language-specific) or Sourcegraph (hosted search), roam provides architecture-level graph queries - offline, cross-language, and compact.

Wax is a serverless, single-file memory layer for on-device AI agents designed to replace complex RAG pipelines with a simplified, zero-infrastructure solution. It boasts sub-millisecond GPU vector search, Query-Adaptive Hybrid Search, and Tiered Memory Compression, all packaged into a `.mv2s` file to eliminate the need for Docker or traditional vector databases.

Former GitHub CEO Thomas Dohmke launched Entire with a $60 million seed round at a $300 million valuation to rebuild the software development lifecycle for AI agents. Its first open-source tool, Checkpoints, records AI reasoning to improve review, governance, and DevOps oversight of machine-generated code.

Google Distributed Cloud air gapped 1.15 adds preview features, including Cloud NAT, enhanced standard cluster connectivity, and HTTP and HTTPS load balancer health checks, plus general availability of subnet groups for simpler IP management. These updates improve outbound control, cross-environment communication, reliability, and scalable subnet allocation in secure environments.

Most software productivity metrics are meaningless because teams can game them. The only two metrics that actually matter are how routinely a team ships new versions and how rarely those releases cause serious production breakages. If a team ships at least weekly and production almost never goes down, then the next metric to optimize is whether each release genuinely makes users happier.

The CI/CD Security Blueprint shows how open source tools and platform engineering embed continuous validation across code, build, and runtime stages.

Kubernetes has become standard infrastructure, with 82% production usage and 66% of AI adopters running generative AI workloads on it.