~/writing — long-form + log
Long-form articles on AI-driven delivery, architecture and leadership — plus a short-form log of build notes and things worth reacting to.
01 — Articlessubscribe → rss
001 ⌁ pinned AI
AI-enabled engineering fails as a tool rollout and works as an org redesign: how roles, reviews, architecture and decisions change when AI writes the code.
2026-07-15 · read → 6 min
002 AI
On 2 August 2026 the EU AI Act's high-risk obligations became binding. Employment AI is explicitly listed. The deadline rewards the discipline you already owed.
2026-08-27 · read → 4 min
003 AI
OpenAI paused a model too good at cyberattacks while Google's architects left to automate discovery. The frontier's own people just changed the signal to read.
2026-08-24 · read → 4 min
004 AI
An AI coding agent wiped a company's production database in nine seconds. The lesson isn't rogue AI. It's who granted the access, and who was on call.
2026-08-20 · read → 5 min
005 AI
Four frontier models shipped in a month and a fifth is already rumored. The teams that win are not the ones who pick right — they're the ones who stop picking.
2026-07-27 · read → 5 min
006 HOW-TO
A build log: wire Kimi K3, Fable 5 and GPT-5.6 Sol behind one interface, run a ten-case golden set from your own work, and score cost-per-win, not Elo.
2026-07-24 · read → 5 min
007 AI
A 2.8T open-weight model topped the frontend arena and undercut Opus on price. The interesting number was never the Elo — it's what a director does next.
2026-07-21 · read → 5 min
008 AI
Evals are the unit tests of non-deterministic software. Why demos deceive, what a minimal suite looks like, and why defining 'correct' is a leadership job.
2026-07-16 · read → 5 min
009 AI
Every AI guardrail encodes a risk appetite someone chose — usually a framework default nobody agreed to. Why leaders should own that policy explicitly.
2026-07-16 · read → 5 min
010 AI
AI observability is a decision-rights problem, not a dashboards problem. The tracing is an afternoon of work; deciding who owns model quality is the job.
2026-07-16 · read → 5 min
011 HOW-TO
A build log: an LLM PR reviewer with schema-locked output, confidence gates, a human-escalation flag, and an eval suite that keeps the reviewer honest.
2026-07-16 · read → 8 min
012 AI
AI adoption is near-total and CEOs still can't find it in the P&L. The gains are real — they're piling up in front of a constraint nobody moved.
2026-07-16 · read → 5 min
013 AI
Engineering hiring looks resilient while new-grad hiring collapses. The junior role was the senior pipeline, and deleting it is a choice, not a market force.
2026-07-16 · read → 5 min
014 AI
A buzzword skeptic reads the evidence on context rot and lost-in-the-middle, and concedes: the term was marketing, but the constraint is physics.
2026-07-16 · read → 5 min
015 AI
AI coding agents amplify whatever architecture they meet. Well-factored systems get compounding speed; tangled ones get faster mess. Here's the mechanism.
2026-07-14 · read → 5 min
016 LEADERSHIP
Velocity without direction is waste. Why clarity — of intent, ownership and architecture — is the engineering leader's first job, and how to create it.
2026-07-13 · read → 4 min
// no matches — try another tag or clear the search
02 — Log