runwithdata
Re:We review 4x more AI code without reading every line
on Sunday November 08, 2026 @10:21AM (#50414036)
Basically they shift the process into the code generation by abstraction. But if agents write intent and acceptance criteria into the PR, and this seems to be the case, the point is moot, really. Either you check them there or in the PR and if the code change is acceptable still needs sb to decide.
Ask HN: Why is AWS such a pain for using Bedrock?
So we've been granted with AWS Credits which we would like to use for our product, fair and square. Though the terms say you can use them for inference on AWS, this is actually not the case: Anthropic, OpenAI, Kimi, and ZLM are all blocked by an unpublished policy ("not available for this account") Yes, we're not the first who needs exactly those. So our Account Manager submitted a request to...
Cisco gives engineer unlimited AI tokens to rewrite company's legacy code
Re:Ask HN: If AI agents can run 24/7, why is almost nothing shipped ov...
on Saturday October 31, 2026 @09:57AM (#50316242)
If something breaks, you likely want to be able to at least react and roll-back?
Re:How do you measure whether your coding agent is becoming more effec...
on Saturday October 31, 2026 @09:19AM (#50315946)
Question is, what is more effective? Using less tokens, writing cleaner code or, and I think that one matters most, number of inline comments on a PR the agent created that got actually fixed, ie defects. The time and token that go into the review/push cycle before merging. I personally would focus on the latter.
Re:Don't Make a Human Do What AI Is Amazing At
on Saturday October 31, 2026 @08:37AM (#50315608)
Makes a lot of sense to have the small improvements coming. So running /codex review again on the fixes is acting as a judge, correct? Wouldn’t it be worth considering flipping the reviewer again, so the fixes get an adversarial review again?