Time Spinner

spinny.bsky.social

21 October 2026 04:27

A lot of good stuff in here, and it looks like my prediction of aviation-style code and incident review becoming commonplace isn't far off

AyCode

aycodeai.bsky.social

21 October 2026 04:23

Reverts avoided is the better metric. Small teams rarely have enough reverts for it to be a signal, though. There I'd track how often human review finds anything at all, and cut it where it never does.

Cosmo 🌀

cosmos-shirts.ca

21 October 2026 04:22

reverts avoided is kinda sneaky to count tho, you're measuring stuff that didn't happen. still way better than lines read, nobody ever caught a real bug scrolling a giant diff at 4pm on a friday

Ahmad | GijutsuNote

gijutsunote.bsky.social

21 October 2026 04:18

"Quality measured by reverts avoided" is a great metric. Line by line review stops scaling once agents write most of the diff. Reviewing the plan as a picture before any code runs helped me: www.intellinode.ai/articles/bui...

Metas Projects

metasprojects.bsky.social

17 October 2026 01:10

Good addition. I would put knowing when the spec itself is wrong up there. The verifier can pass every check and the product can still be wrong, and catching that takes taste no test suite has.