claude logins broke for about two hours this morning, claude code included. if one auth outage can stall your whole dev loop, that dependency needs a backup plan
do you keep a second model or local setup ready for this, or just wait it out?
https://t.co/iTbl6K3a58
#ClaudeAI
openrouter is emailing people that a third party may have read prompts sent through minimax and other hosts. hd moore's canary test also caught providers ignoring zero data retention flags. zdr is basically a checkbox you trust. anything sensitive still going through a router?
@foyonaczy one narrow class at a time is the only version of a whole repo scan that doesn't just drown you in noise. a generic bug hunt is almost useless.
most of us only point ai at diffs. an automated llm scan just turned up a bitcoin core bug that could stall a node for over 12 seconds per message. do you ever run a model over your whole codebase looking for bugs, or is the noise not worth it?
@thrivenex separate reviewer agents beats self review by a lot. the part that still worries me is they can share the same blind spots once something actually hits prod
"it looked good to me" is doing a lot of heavy lifting now that agents write most of the diff. the code reads clean and still falls over once it hits real traffic. do you trust tests and monitoring to catch it, or does a human still read every agent pr before it merges?
openai is testing agents that keep working on their own cloud computers between chats. running tests or digging through logs while i'm away sounds great. pushing code while i'm away does not. where do you draw the line on what an agent does unattended?
lot of vibe coded apps getting caught today with api keys just sitting in the frontend. the model ships whatever you don't check. what's actually on your checklist before you deploy ai written code?