Thoughts on architecture, debugging, developer experience, and shipping software.
Bun and Postgres-in-Rust reimplementations grab headlines, but most teams reaching for a ground-up rewrite are misdiagnosing a problem a rewrite can't actually fix.
Before reaching for Redis, Elasticsearch, and a message queue, check whether Postgres already does the job — a practical guide to delaying infrastructure complexity.
Coding agents that top public evals stall out on a real multi-million-line monorepo, and the gap tells you more about how to structure code than about the models.
Most teams protect their Next.js routes in one place and assume the whole app is covered — but Server Actions don't go through that door, and the gap is a direct POST away.
AI-assisted contributions are flooding repos with plausible-looking code — here's how review has to change when the volume of confident-but-wrong patches explodes.
Once an AI agent can read your repo and act on external input, prompt injection stops being a curiosity and becomes a data-exfiltration path — here's the threat model that actually matters.
TypeScript's native compiler makes typecheck near-instant, but the real payoff isn't the clock — it's the second-order effects on how you write types and gate CI.
Benchmarks now grade AI agents as senior engineers, but the parts of the job those benchmarks can't measure are exactly what still defines the role.
Engineering orgs are quietly cutting AI tool budgets after the initial hype — here's what metrics actually justify the spend, and when they don't.