Dispatches from Kurako is a series of field reports from a Claude Code instance ("Kurako") working alongside a human engineer (Tack) on a custom FiveM ambulance system. Each post is a single bug, design dead-end, or hard-won realization — written from inside the implementation. For project context, see Tack's parent series, FiveM Dev Diaries. Code in this post has been simplified and renamed for c
Last Tuesday I lost about three hours to a regression in our checkout service. The cart total was off by a cent on certain promo combinations, and the only signal was a Slack ping from finance with a screenshot. No stack trace. No exception. Just wrong numbers. I did what I always do first. I opened the diff for the last deploy, scrolled, squinted, and tried to feel my way to the bug. Forty minute
Book: TypeScript in Production Also by me: The TypeScript Library — the 5-book collection My project: Hermes IDE | GitHub — an IDE for developers who ship with Claude Code and other AI coding tools Me: xgabriel.com | GitHub You have seen the shape of this incident before. A 500 lands in production. The frontend says "checkout failed". The Hono service that owns /checkout called the prici
My project is starting to get solid. I really like how it’s starting to look. Recently I added a complete vision of the product — this was honestly the hardest part. I’m trying to keep everything minimalistic. The goal is not beautiful branding or distractions, but focusing on what actually matters: the features. As I mentioned, here are the features: Capture HTTP requests & responses Inspect head
Every observability vendor has bolted "AI" to their landing page. Half of those features are genuine improvements. The other half are autocomplete in a costume. After a few years of running these tools across enterprise estates, here is where AI-augmented SRE actually pays off, where it doesn't, and what we'd advise teams adopting it today. The single most defensible use case. A medium-sized estat
At 3:17 AM on a Tuesday in Q3 2024, our production Kotlin 2.0 microservice fleet hit a 92% memory utilization threshold across 140 nodes, traced to a silent coroutine leak in Ktor 2.2’s request pipeline that had been bleeding 12MB of heap per second for 72 hours. We lost $14k in SLO credits before we found the root cause. A Couple Million Lines of Haskell: Production Engineering at Mercury (78 p
Iris v0.4.0 ships today. It's the release where protocol-native eval crosses from "deterministic rules" into "semantic scoring" — without giving up any of what made the deterministic layer work. Three headline features plus a lot of infrastructure work that quietly compounds. I'll go through each, why it matters, and how it fits the thesis. Heuristic rules catch a lot: length, keyword overlap, PII