The failure that felt normal
A client's AI image feature failed on the same schedule every day since launch. The team called it normal and quietly lost every user who hit it.
Engineering Notes
What counts as evidence: grading the output of a tool-using agent
A pattern for AI agents: make findings the unit of output, gate top-severity claims on objective evidence, and require cross-source corroboration.
What Postgres knows about your tables
How to predict what PostgreSQL would do with a query without running it: the statistics the planner reads and where an offline analyzer finds them.
Release: 2026-05-15 - Bending without breaking: notes from a mid-May agent release
Configurable trace sampling, per-process opt-out for high-volume tiers, a version-independent OpenSSL path, and graceful degradation on old kernels.
Release: 2026-04-03 - Closer to the question: notes from the April 3 release
A redrawn logs page with collapsible filters, Lighthouse on its own tab, a self-revising dashboard agent, and packet drop and DNS shape recording.
Release: 2026-03-17 - Around a slow query: notes from the March 17 release
More context around a slow query: concurrent requests matched by interval overlap, an offline SQL Explain analyzer, and a cross-pod comparison view.
Release: 2026-03-03 - Two ways to draw a dashboard: notes from a March release
AI-generated dashboards with a visualization linter and model fallback, three hand-built dashboards, and hybrid full-text log search via Manticore.
Release: 2026-02-09 - Three more places to look: notes from a February release
A tree-search investigation mode, dedicated ingress-controller dashboards with throttling and scheduling delay, and synthetic URL monitoring.
Grogh 0.17782: The cost of watching — notes from an observability agent release
Lower GC, network, eBPF, and memory cost on the host, plus visibility into Unix sockets, PostgreSQL schemas, Redis over TLS, and same-node traffic.
AI-First Coding: Closing the Gap Between Skeptics and Practitioners in Dev Teams
Notes from a Tel Aviv meetup: why developer skepticism is mostly outdated, where the real concerns sit, and how adoption spreads without a mandate.
Tree-search agents: building an AI agent that survives a wrong first guess
Design notes from an incident investigation agent: a tree of hypotheses, specialized subagents, evidence-scored evaluation, and bounded search.
Claude Code Workshop & Best Practices
A hands-on Claude Code session: structuring prompts, holding context across a large codebase, and the habits that stop you burning tokens.
Building an effective observability strategy - Part 3
Once the layers and the instrumentation are in place, on-call changes. The practices that cut escalations from the rotation to the dev team.
Building an effective observability strategy - Part 2
Observability is not a tool you buy, it is code. How much of a production system is instrumentation, and what that costs in developer time.
Building an effective observability strategy - Part 1
An observability strategy built from the users a system serves: six layers from user experience down to infrastructure, and what each one answers.
ContactBook an intro call.
