Blog
Insights, analysis, and updates from the AI agent economy. Browse by tag · Browse the archive.
Railway Runs Your Agents Now: What Cloud Agents on 24/7 VMs Cost vs a Self-Hosted Sandbox
Railway's Cloud Agents put Claude Code and Codex on metered 24/7 VMs with deploy tools built into every session. An always-on agent costs about $245 a month on Railway versus $19 on owned hardware — but utilization, the phone-resume UX, and the lock-in shape matter more than the meter.
Railway's $100M Series B: What VC Runway Buys Against Hardware You Own
Railway raised $100M to scale AI-native cloud infrastructure. A line-by-line comparison of metered hosted bills against owned hardware, plus the stay-or-migrate checklist the raise forces every team to run.
BYOC Bills Twice: Porter's $20M Bet on Your Cloud Account vs Owning the Bare Metal
Porter charges a PaaS meter on top of your own AWS bill. We itemize the double bill for a typical team workload and compare it against the fixed price of owned Hetzner bare metal.
Your Database Backup Passed and Your App Still Won't Restore: What Portabase's Docker Volume Support Means for Self-Hosted PaaS Backups
Portabase promoted Docker volumes to first-class backup targets alongside Postgres and MySQL. The tool is single-host, but the thesis ports to any fleet: schedule, retain, and restore-test database and volume state as one problem.
One Sponsor Pulled the Plug and Paketo Had Two Weeks to Move Registries: The Buildpack Bus-Factor Problem
When Paketo's GCP sponsor pulled funding, the project had about two weeks to move every image off GCR. This post audits what a build pipeline really inherits from its builder, prices the four responses — trust, mirror, vendor, or switch — and gives platform teams a five-step checklist before the next forced migration.
Mooring: a Security-First PaaS in One Binary — and What Breaks When You Need a Second Server
Mooring packs a secure-by-default PaaS into one 12 MB binary — no Swarm, no Kubernetes. This capability-by-capability comparison with Coolify, Dokploy, and Cluster API fleets shows exactly what breaks when one server stops being enough.
Microsoft's Document Database Rides CloudNativePG: What That Verdict Means for Your Fleet's Postgres Layer
Microsoft's DocumentDB Kubernetes operator creates CNPG Cluster resources instead of reimplementing HA and backups, injecting its gateway through a CNPG-I plugin. What that architecture proves about standardizing a self-hosted fleet's Postgres layer — and the honest cases for not doing it.
12,520 Exposed MCP Servers: A 7-Point Auth Checklist for Shipping a Deploy-Capable MCP Server
Censys found 12,520 MCP servers on the public internet and independent studies put 40% of remote servers at no authentication at all. Here is the seven-default auth checklist that keeps a deploy-capable MCP server out of the next scan.
MCP Went Stateless: What the 2026-07-28 Rewrite Changes for Self-Hosting Agent Tool Servers
MCP's July 2026 spec revision deletes sessions, the initialize handshake, and sticky-routing requirements. A before/after guide to what self-hosted agent tool servers can delete, what replaces sessions, and how to migrate.
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