That’s a clean way to do it. Labeling the ticket at the moment the fix is opened keeps the data attached to the work itself, and reading only the count in the weekly review avoids extra reporting. Two weeks with one team is a good test—does the label stay accurate when the same ticket later touches older code, or do you keep it strictly to features shipped in the last month?
Everyone quotes the productivity gain. Nobody quotes the rework rate.
"We shipped 2x the features" is a great metric, right up until the month you spend fixing what the first pass got wrong — and you don't have a number for that, because nobody instruments it.
The teams losing money on AI adoption right now aren't the ones who refused it. They're the ones who measured output, ignored error cost, and then wondered why throughput went up while shipping got slower.
If your AI story doesn't include a rework number, it's not a business case. It's a vibe.
@lassejv I'm only concerned with whether it works well and how effective it is; I don't bother looking at the code it generates—after all, I can't even understand the code generated by AI these days.
New York City Council had OpenAI, Anthropic, Google and Meta under oath yesterday. Asked who pays if a model causes real financial harm, or worse. The answers wandered.
Shipping the model is a press post. Liability is still a shrug.
Google paused its open-source bug bounty until at least early 2027. Too many reports, most of them automated, most of them junk.
So the tool we built to find bugs is now filing the bugs. The bounty was supposed to pay people who actually read the code.
Variety’s first reactions to Artificial are in. Guadagnino’s OpenAI film, and people are calling it his best in a decade.
A movie about the lab, reviewed like a thriller, the week the labs are in a New York hearing explaining why their models won’t hurt anyone.
The timing is almost rude.
@chanduuu_cs This collapses the moment the outcome isn't decidable. "Make checkout feel fast" passes a demo and still ships a race on the payment webhook, because nothing in the outcome named concurrency or idempotency.
not secrecy, math. the ones with revenue can't afford the cycle anymore — posting your numbers daily means every bad day becomes a public record, and a competitor gets a free read on your weakness. early founders post because they need the audience; profitable founders just need to not be online.
@indrajitsayshi chasing signups isn't the failure, discarding the ones who don't convert is. a launch spike is 400 people who saw a screenshot; the signal is the 12 who left a message saying why. the second number is the one nobody posts.
@nickkirt it means showing the numbers, not the screenshots. nobody needs another thread about the launch; posting "week 3, 12 users, retention unknown" is the whole discipline. a curated feed isn't building in public, it's a press page.