Metrics went from telling you something broke to deciding which GPU handles your inference request. The pattern hasn’t changed in 20 years, we just got better at acting on the data.
Making an NVIDIA eGPU Actually Work on Linux (The Hard Way)
Debugging an NVIDIA RTX 3070 eGPU that refused to work on a Framework 13 with Arch Linux — PCI BAR allocation, Thunderbolt bridge windows, and getting KDE Wayland to composite on the right GPU.
Maslow’s Hammer and the MCP Debate
MCP has been called ‘a layer of unnecessary indirection.’ It’s not — but it’s not a universal solution either. Here’s when MCP makes sense, when a REST API is enough, and why the real value is in a place nobody talks about.
Kill Your Service Account Keys: Secure GitLab CI/CD on Google Cloud
How I built a fully keyless CI/CD pipeline from GitLab to Google Cloud — with Workload Identity Federation, Binary Authorization, vulnerability scanning, and progressive delivery. No service account keys were harmed.
Four People, Four Datacenters, Three Thousand Servers
How we managed 3,000 servers across 4 datacenters with a team of four — and what I learned about automation, constraints, and doing more with less.