The Journal

Field notes from a career of building systems that scale.

Architecture, machine intelligence, engineering leadership — written for technical leaders who'd rather build than speculate.

Filter

ai · SEP 10 · 2026

Just let Claude Code write your software factory

I expected building a software factory orchestrator to take weekends of framework evaluation. One prompt to Claude Code produced a working pipeline in an hour — and it caught two design flaws before the first run.

#ai #claude-code #agents #howto Read essay →

ai · SEP 01 · 2026

Six Months In: Where the 2026 CTO Bets Actually Stand

Half a year on from the three predictions — orchestration, enterprise enablement, everyone builds — here is what landed, what stalled, and the talent risk nobody is talking about.

#ai #agents #cto Read essay →

ai · AUG 31 · 2026

Stop trying to build the perfect AI harness

Six months ago everyone was tuning CLAUDE.md, AGENTS.md, prompt libraries, and skill catalogues like the model was a junior that needed instructions taped to every wall. The harness is starting to look like yesterday's fix for a model that's already moved on.

#ai #agents #agent-skills Read essay →

ai · AUG 23 · 2026

1,227 Requests in 191 Days: What Lighthouse's Agentic Score Actually Measures

Chrome shipped a Lighthouse category in May that scores whether you publish an llms.txt. Google's own AI guide says AI crawlers ignore it. Both pages are still up. The audit is real, the score is nonsense, and the actionable audits (CLS, a11y) are ones we already had.

#ai #agents #seo Read essay →

ai · AUG 06 · 2026

TIME Built an AI-Only Website. My Blog Already Had One — Minus the Ads

TIME.com serves AI crawlers a stripped markdown version of every article, with ads no human ever sees. This blog has served machine-readable markdown twins for a while — today I closed the last gap, in fifteen lines of .htaccess.

#ai #agents Read essay →

wasm · JUL 15 · 2026

5 Reasons Your Next Deploy Should Be 145 KB of Wasm, Not Docker

I replaced this blog's web server with a single 145 KB WebAssembly binary on Cloudflare Workers — smaller than most images on this page, $0 a month, and tested locally with the exact artifact that ships. Here's why that combination beats a container for a whole class of services.

#wasm #howto Read essay →

claude-code · JUL 14 · 2026

Can a Claude Skill Out-Design You? I Handed It My Blog to Find Out

I installed the Hallmark design skill in Claude Code and told it to redesign my blog. Forty minutes later it had beaten a design I'd sweated over with Google Stitch and Claude Design.

#claude-code #agent-skills #design #ai #howto Read essay →

ai · JUL 12 · 2026

How I Made a Minimalist Agent Harness Code Like a Senior Engineer

I pointed a bare-bones agent harness at a coding model, bolted on a pack of best-practice engineering skills, and had it spec, build, and review a Matrix-themed snake game. The code it produced could have come out of Claude Code or any other harness I already use.

#ai #agents #pi #agent-skills Read essay →

ai · JUL 09 · 2026

How I Made My Blog AI-Readable Before My Claude Bill Goes Up

I pointed Fable 5 at a website specification, told it to audit my blog, plan the fixes, and ship them. Here's how I got the site up to spec—and machine-readable—in one sitting.

#ai Read essay →

ai · JUN 30 · 2026

Fast GLM 5.2 Plans, Cheap Models Code: One LiteLLM File for Claude Code

Claude Code already splits work across model tiers — a heavy thinker and cheaper background models. So point the thinking tier at a fast GLM 5.2 and the rest at pennies-per-token providers. One LiteLLM config file does it, and Claude Code never knows the difference.

#ai #claude-code #howto Read essay →