The Draft: Issue 43 — Your Agents Are Shipping. Who's Reviewing?
Your Agents Are Shipping. Who's Reviewing?
🌟 Editor's Note
AI agents are writing infrastructure code faster than most teams can review it. That's the quiet thread running through the community this week — not whether agents are useful, but whether humans are keeping up with what's being shipped. OpenTelemetry graduated at CNCF, teams are fleeing hyperscalers for bare metal, and Kubernetes storage remains the thing nobody wants to touch in production. This week: speed versus accountability.
🚀 Job Market & Career Outlook
The DevOps role keeps splintering — platform engineer, cloud engineer, AI cloud engineer — but the core skill set hasn't changed as much as the job boards suggest. What has changed: employers want hybrids who can write the automation, own the architecture, and talk cost to the business. That's making the entry wall brutally high. Junior DevOps is functionally dead as an entry path.
What's landing interviews: hands-on proof. Build something that deploys, breaks under load, and produces a postmortem worth reading. That artifact beats any certification. Burnout is a recurring theme too — roles are quietly shifting from "build infrastructure" to "review agent output," and nobody renegotiated the expectations.
🤖 AI Agents & Practical Automation
The post that captured the week: "After 3 months building my personal AI assistant, I think hype > reality." Fifty-plus comments and a clear consensus — agents work for bounded tasks, but the "replace your team" narrative is collapsing under real-world experience. The sharper problem: "My job shifted from writing code to reviewing it, and I don't remember agreeing to that."
A new class of agent governance tools is emerging — firewalls for what agents can do, audit trails for what they did. The community consensus: "Dashboards aren't governance. Enforcement must live between agent and infra." If you can't reconstruct why the agent made a specific choice, you're not reviewing. You're rubber-stamping with extra steps.
💻 Coding Corner / Practical Implementation
Previewuse spins up per-branch preview environments using Docker Compose — every PR gets its own isolated deployment. Warool manages reverse SSH tunnels for exposing local services behind NAT, eliminating ngrok dependencies. Sentinello monitors CVEs against your dependency tree and alerts before they hit production. All open source, all solving friction that most teams tolerate until the audit.
🧯 Infrastructure Pain Points
The story that stopped people scrolling: a team migrated from Azure at $7,900/month to Hetzner bare metal plus Cloudflare plus GitHub Actions — for $330/month. The caveat matters: you need the ops skill, and compliance gets harder without managed services. But the bill is real.
Terraform state bloat is back in the pain rotation — mono-repo sprawl, failed workspace scaling, drift nobody noticed until a plan showed 200 unexpected changes. Community prescriptions: modularize aggressively, compartmentalize state, use Terragrunt, add drift detection. And Kubernetes persistent storage? Still the part of production that teams actively avoid touching.
🔦 Tool Spotlight — Envoy Gateway
What it is: An open-source implementation of the Kubernetes Gateway API, built on the Envoy proxy. Managed control plane for routing, TLS, and traffic management without locking into a specific ingress controller.
Why it matters: Gateway API is replacing the Ingress resource as the routing standard, and Envoy Gateway is the community's leading implementation. The key win: clean separation between Gateway ownership (infra team) and Route ownership (app teams). That split solves an organizational pain point Ingress never addressed.
- Best for: Multi-team K8s clusters needing separate routing ownership
- Watch out for: Complex rate limiting may still require lower-level Envoy config
- Verdict: If you're starting a new cluster or planning an Ingress migration, this is where the ecosystem is heading
"The separation between gateway and route ownership is… the biggest improvement." — community consensus on Gateway API adoption
📈 Emerging Trends & Updates Generating Buzz
OpenTelemetry graduated at CNCF. Graduation signals production readiness to procurement teams. For practitioners, it means one instrumentation layer across traces, metrics, and logs. If you've been holding off because the API was "still incubating," that excuse is gone.
Self-hosted infrastructure is surging at small and mid-company scale. Local AWS and Azure emulators, Hetzner, Cloudflare — teams are voting with their wallets. And cost governance is becoming a first-class discipline — per-workflow cost attribution (especially for AI API spend like AWS Bedrock), hard budget alarms, and spending governance. If you can't measure cost at the workflow level, you can't control it.
