Articles on reliability, leadership, and platform engineering. Lessons from the field and late-night debugging sessions.
23 articles · 92 tags · 72,411 words
Interactive simulations to practice incident response and observability skills.
Experience a real 2AM incident response: golden signals, distributed traces, log correlation.
Respond to incident →Black Friday checkout outage scenario. Make decisions, get scored on your SRE skills.
Start simulation →Interactive modules on observability and resilience. Percentiles, SLOs, chaos engineering, and more.
Start learning →Filter by tag
Researchers found that nine popular AI chat services contacted advertising or tracking companies, sometimes exposing chat titles, links, messages, or screenshots. The leak may have started as a design flaw, but the business model gives providers little reason to remove it.
Read articleKubernetes 1.35 made CPU and memory requests mutable on a running pod. CPU can use that. A JVM heap cannot. The resize policy, VPA mode, and HPA collision that follow from that.
Read articleSkills, agents, loops, and graphs are not a maturity ladder. They're four small, mostly old ideas with new names. Here's what each one actually means, what transfers from ops work, and where the hype is louder than the substance.
Read articleOTEL looks like a vendor deck until you see the actual pipeline: apps talk to a node agent, the agent talks to a cluster gateway, traces land in Datadog, and logs land in Splunk. One instrumentation path, two exporter mappings.
Read articleA prompt is one instruction you have to be awake to send. A loop is a standing agent that holds your context, chases a goal, and keeps going while you sleep. You already build control loops for everything that matters in production. Here is how to point that instinct at your agents: loop anatomy, shared state files, drift prevention, hard gates, the autonomy ladder, and five platform fleets worth standing up this week.
Read articleEvery AI feature you ship runs a meter you can't see on the dashboard. It shows up as a token bill that climbs faster than usage and answers that quietly get worse. The reflex is to blame the model and swap in a cheaper one. That fixes nothing. The charge comes from the harness you built around it, collected on every turn. Here's where it comes from and how to stop feeding it.
Read articleMost engineering onboarding fails not because the documentation is bad but because documentation was never the right tool for the job. Here's the architecture that retrieves the connected trail of decisions instead.
Read articleSomewhere between 'we have no visibility' and 'we have too much data to make sense of,' something went wrong. This is about what that something is and how the signals your systems emit can either answer the health question or make it impossible and make you go bonkers on a 2AM page.
Read articleCompanies are buying AI tools for coding, operations, support, and data analysis. The real question is whether the token spend is turning into better work, measurable outcomes, and OKRs the business can defend.
Read articleGuardrails, secure code, environment segregation, and the review patterns we built for a healthcare AI platform under HIPAA and SOC 2 scrutiny. No fluff. Just what we actually shipped.
Read article