Blog
Insights, analysis, and updates from the AI agent economy. Browse by tag · Browse the archive.
Railway Cloud Agents Promote Your Dev VM Straight to Prod: What Skipping the Build Actually Costs
Railway's Cloud Agents Beta promotes the agent's lived-in VM straight to production via a VM daemon, skipping git push and CI as too slow. A five-row ledger of what the build step was quietly doing — drift, secrets, approvals, rollback, provenance — and how to keep the fast loop without shipping a mutable machine.
Every Pull Request Gets Its Own Environment: The Real Build-vs-Buy Math for Preview Environments
Shared staging lets a team finish 12 of 42 tickets while the rest queue. Preview-per-PR fixes that — the question is whether to build it on a Cluster-API fleet or buy it from Bunnyshell, Uffizzi, or Northflank. First-year totals, the breakeven near 20 concurrent PRs, and the six jobs every solution must cover.
Docker Is Enough Until the Second Server: Why We Still Bet on Kubernetes
2026's PaaS-first chorus says Docker on one box handles 95% of workloads and Kubernetes is overkill. The verdict table, the Coolify/Dokploy/CapRover facts, and the five things that break at the second server — plus where the backlash is completely right.
k3s Now, EKS Later? Pricing Ownkube's Graduation Ladder Against Cluster API on Day One
Ownkube lets teams start on hosted compute and graduate the same app into their own AWS account, from free single-node to metered multi-AZ. A priced run of one steady 2-vCPU team workload across every rung — plus what re-surfaces on migration day — against committing to Cluster API on flat hardware from the first deploy.
10,000 MCP Servers and No Human Wiring Them Up: What Server Cards and Stateless Operation Mean for Agents That Deploy Your Apps
MCP passed 10,000 public servers and is going stateless and self-describing, so agents will soon reach your deploy tools with no human wiring anything up. Here is the three-tier tool surface — observe freely, gate mutation, deny destruction — that survives contact with autonomous callers.
Your Coding Agent Needs a Sandbox, Not a StatefulSet: gVisor vs Kata Under Kubernetes' New Sandbox Controller
Kubernetes SIG Apps shipped an upstream Sandbox controller for running untrusted agent code. A practical decision guide: what it replaces, whether gVisor or Kata fits your tenants, and what warm pools cost in idle capacity.
Kubernetes 2.0 on HN: Why Rivet's Docker Compose Story Ends at the Second Server
A skeptic on HN asked how Docker Compose could be a more reliable, scalable alternative to Kubernetes for Rivet's actor platform. Rivet's own docs supply the answer: the Postgres-plus-NATS second-machine path, the 1,000-actor ceiling, and where K0s fits instead.
Kubernetes 1.37 Will Reject Your Static Pods: the kubeadm Audit to Run Before You Upgrade
Kubernetes 1.37 removes the opt-out and has the kubelet refuse any static pod referencing a Secret or ConfigMap. The one-line audit for every control-plane node, the full list of rejected fields, and the hostPath and extraVolumes patterns that replace them.
Your GPUs Are 60% Idle and Kubernetes Says Everything Is Fine
A July 2026 CNCF case study found Kubeflow GPUs ~60% idle with every pod green — a scheduler-vs-network conflict fixed with topology constraints (40% to 85% utilization). The three families of GPU waste, a DCGM-based instrumentation ladder, and the rule for earning your next GPU node.
Subscribe
New posts land in your reader as soon as they publish. Pick a format — all three carry the same posts.
Current feeds keep roughly two days of posts so daily polling does not miss a burst. Older entries stay reachable from the feed's next-page link in readers that follow it, or from the blog archive.
Following one topic instead? Browse tags