§ WRITING
Reasoning made visible
Once agents write most of the code, the signal isn't volume — it's judgment. Evals, security reviews, and proof that what shipped actually works.
Deliver code you've proven to work
Manual testing, automated tests that fail on revert, and a narrated security review of agent-written code — reconstructing a class of authz bug I caught, generically.
Evals in public
Take a real task, build a small eval suite, show the failure modes it caught. Turning a crappy eval into a good one — the 2026 non-negotiable.
[PLACEHOLDER: THIRD ESSAY]
[PLACEHOLDER: A PROBLEM→DEPLOYMENT NARRATIVE, OR "HOW I DIRECT CODING AGENTS SAFELY" — WHICHEVER YOU WRITE FIRST.]
READY-TO-FILL · TITLES & TOPICS DRAFTED FROM THE STRATEGY DOC · SWAP DATES / STATUS AND FILL EACH POST AS YOU PUBLISH · ALSO SURFACE THESE IN LINKEDIN FEATURED + ON X