The Draft

Issue 47  ·  DevOps & Platform Engineering Intelligence

The Boring Infrastructure That Actually Holds

Week of June 29, 2026

🌟 Editor's Note

This week the community handed me a thesis statement I didn't have to write myself: the orgs that are actually winning right now aren't the ones chasing the latest agentic framework. They're the ones that got boring infra right first — IaC coverage, drift detection, tagged resources, documented runbooks — and then layered AI on top. The rest are building on sand and hoping nobody notices. Someone always notices.

🚀 Job Market & Career Outlook

The market is brutal and the internet is not lying to you about it. Thirty-five-plus active threads this week on layoffs, hiring slowdowns, and skill mismatches — and the recurring theme is frustration from people who've done the work. They upskilled. They got the certs. They're still waiting on callbacks. “4 jobs in 4 years: 2 layoffs, 1 bad fit” is not a resume problem. It's a market problem.

What is getting attention in postings: Terraform, Python, Kubernetes, GitHub Actions, AWS/Azure automation, and — increasingly — demonstrated comfort with AI tooling. Not “used ChatGPT for a side project” comfort. Actual integration of LLMs into incident response, cost reporting, and CI/CD workflows. The bar moved and a lot of people are still training for the old bar.

One practical note that surfaced repeatedly: resume keyword matching matters more right now than it should. Before you apply, run your resume against the specific job description and close the gap. It's an imperfect game but it's the game being played.

🤖 AI Agents & Practical Automation

The most-upvoted post this week: “Fed 5 days of k8s logs into a 1M context model — found the root cause our team spent 14 hours on in about 90 seconds.” The thread has 80+ comments and the signal-to-noise ratio is unusually high. People are genuinely seeing results from AI-assisted incident analysis — cross-referencing logs, metrics, Slack threads, and past runbooks simultaneously in ways that aren't humanly efficient at 2am.

But the community's skepticism is equally well-represented. The best counter-comment: “AI can be really good at finding signal in huge amounts of data. It can also be really good at hallucinating signal in huge amounts of data.” That's the whole problem in one sentence. The use cases that actually work are where the AI is assistive — surfacing hypotheses, narrowing the search space — not where it's authoritative.

The emerging pattern from practitioners who've integrated this well: deterministic recovery workflows first, agentic layers on top. Not the other way around.

💻 Coding Corner / Practical Implementation

A few things worth bookmarking this week. On the Kubernetes cost side, a post on 5 EKS changes that generated real savings got traction — VPA in recommend-only mode, Karpenter consolidation, and aggressive right-sizing of resource requests. The VPA tip is worth repeating: never run it in auto mode, use it to generate recommendations and apply them deliberately.

Terraform Permadiff also showed up this week — an open-source CLI for analyzing plan diffs. If you're trying to get a handle on persistent plan noise or catch infrastructure drift in CI, worth a look.

🤯 Infrastructure Pain Points

Drift is still the loudest pain point — 44+ mentions across threads this week. The pattern is always the same: org starts strong with IaC, someone makes a quick console change under pressure, that change doesn't get committed, six months later nobody knows what state the account is actually in. “We're running 9 AWS accounts and maybe half the infra is in Terraform and the other half is just... there.” This is not a rare scenario. It's the default.

The community's hard-won advice: lock down prod accounts so ClickOps is structurally impossible, not just discouraged. Monitor uncovered resources as a metric. And treat your disaster recovery dry runs like SOC2 evidence — because “we think we can” is not a recovery plan.

🔦 Tool Spotlight — Karpenter

Karpenter keeps appearing in cost optimization threads — it's no longer a “should we try this?” conversation. Teams that have shipped it are reporting meaningful savings, particularly when combined with spot instance management and PodDisruptionBudgets wired correctly.

The honest take: Karpenter's consolidation behavior is aggressive by default. That's the feature. But if your applications haven't implemented graceful shutdown properly, you will feel it. Sort out your PDBs and termination handling before you let Karpenter start rebalancing. The savings are real — the pain is avoidable if you prep correctly.

Worth pairing with VPA recommendations and namespace quota enforcement if you're doing a full EKS cost pass.

📈 Emerging Trends & Updates Generating Buzz

The Terraform vs. Pulumi debate had a quantitative edge this week: roughly 18 out of 40+ comments pushed Terraform/OpenTofu, ~10 advocated Pulumi — and mostly for greenfield or strongly developer-centric shops. The consensus: “Terraform is boring and boring is good.” One comment that landed: “The first month of Pulumi feels amazing. The second month is a footgun.” That's not an argument against Pulumi — it's an argument for knowing what you're signing up for.

AWS Lambda MicroVMs are generating genuine curiosity for interactive code environments. Native syslog ingestion to CloudWatch is also live — small feature, useful for teams routing logs through intermediaries. And Headlamp now has a Cluster API plugin, which continues to make it a serious contender for teams wanting a self-hosted, extensible Kubernetes UI.

💬 Quote of the Week

“The difference between ordinary and extraordinary is that little extra.”

— Jimmy Johnson

The Weekly Draft

You're receiving this because you subscribed to The Draft. Unsubscribe