Quick takes between the dispatches —
— short opinion blurbs, written in the margins.
NoteTeaching CLAUDE.md to 500 engineers at EngX
Two days, 500 engineers, a 90-minute talk on teaching Claude what it can't learn from your code. Here's what crystallizing five years of instincts taught me.
APR 21 · 1 minEveryone
NoteClaude's session limits accidentally fixed my burnout
Anthropic tightened session limits last month and quietly dropped default thinking effort. I'm somehow relieved AND frustrated about it.
APR 15 · 2 minEveryone
NoteClaude won't kill junior engineering jobs
I keep hearing this take and I heavily disagree. But the juniors who don't adapt are going to get left behind.
APR 9 · 2 minEveryone
NoteWhen Claude starts hallucinating, kill the session
Most people don't realize they're 30 minutes into arguing with an AI that stopped listening 20 minutes ago. The fix: throw the session away.
APR 8 · 2 minHands-On
NoteDon't let Claude grade its own homework
Most teams think AI code review happens after a PR opens. That's too late. Here's the two-Claude pattern Anthropic uses internally — and how to steal it.
APR 2 · 1 minHands-On
NoteTeaching AI is the fastest way to sharpen your own thinking
Two days, hundreds of engineers, a curriculum months in the making. Teaching forced me to crystallize my own thinking on what it means to work agentically.
MAR 24 · 1 minEveryone