Blog
Insights, analysis, and updates from the AI agent economy. Browse by tag · Browse the archive.
Two European Bare-Metal Giants Diverge: OVHcloud Chases Kubernetes AI Conformance While Hetzner Stays the Self-Hoster's Default
OVHcloud earned CNCF Kubernetes AI conformance in March 2026 while Hetzner kept the price floor. A numbers-first comparison of the badge versus the box — and the workload-shape rule that tells a self-hosted fleet which bet to place.
514 Blueprints vs 354 Apps vs 280+ Services: Why Template Count Is a Vanity Metric That Hides the Single-Box Ceiling
Dokploy lists 514 blueprints, CapRover 354 apps, and Coolify 280+ services. What a template count actually measures — packaged Compose files plus contributor throughput — and the five capabilities no catalog entry provides once one box is not enough.
NETSCOUT Put Network Evidence Behind MCP: What a Vendor-Owned Agent Tool Server Means for Self-Hosted Agent Ops
NETSCOUT's September 2026 MCP launch gives AI agents on-demand access to real-time network evidence. Here is what the vendor-MCP wave means for teams running agents against infrastructure they own: rent the evidence, keep every mutating action on a surface you operate.
Netcup Vouchers vs Hetzner Post-Hike Prices: A Fleet-Role Placement Guide
Hetzner's June hike repriced dedicated cores to €21.50 while netcup's September vouchers cut RS root servers to about €2.24 per core. A role-by-role placement guide for small EU fleets — control plane, elastic workers, batch — plus an honest verdict on second-provider diversification.
Resize the Queue, Not the Pods: Kubernetes 1.36's Mutable Job Resources for Bin-Packing AI Sandboxes
Kubernetes 1.36 promotes mutable pod resources for suspended Jobs to beta: patch a queued job's CPU, memory, or GPU ask to fit free capacity instead of deleting and recreating it. How the suspend-resize-resume flow works, what it buys bursty AI-sandbox pools on owned nodes, and where it still falls short.
Mistral Bought Koyeb: The Tenant Exit Playbook
Koyeb's February 2026 acquisition by Mistral pivoted the platform toward AI inference — this tenant exit playbook rates every Koyeb primitive, from Light Sleep to Serverless Postgres to per-second GPUs, on how cleanly it maps to infrastructure you own, with a move order that keeps each step reversible.
MinIO Vanished From Docker Hub: The One-Line Fix and the Migration You Actually Owe Yourself
MinIO deleted minio/minio and minio/mc from Docker Hub in mid-September 2026, breaking every cold docker compose pull. The quay.io repoint restores green builds today — but that image is frozen forever, so here is the honest SeaweedFS vs Garage vs RustFS comparison plus the registry mirror that makes the next deletion a non-event.
MCP Lock-In Moves Up the Stack: Why Your Infrastructure MCP Server Belongs on Your Own Fleet
MCP passed 10,000 public servers and 97M monthly downloads under foundation governance — but lock-in moved to registries, client defaults, and trust signals. Why teams running deploy-capable infrastructure servers should self-host the server, the catalog, and the registry surface.
MCP's 2026 Roadmap Bets on Stateless Scale: What It Unlocks for a Self-Hosted Deploy Server
MCP's 2026 roadmap puts stateless transport ahead of new features, and the July 2026 spec delivers it: no sessions, no sticky routing, any replica behind an ordinary load balancer. Here is what that unlocks for a self-hosted deploy server — and what is safe to build today versus still landing.
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