Remix.run Logo
▲ tflinton a day ago

I’ve had AI running to auto upgrade our infrastructure and it’s successful twice as much as humans. Failure rate of a human managed upgrade is 1/10, failure rate of an AI agent is 1/20 in my small set data.

Problem is I’ve had 5x more incidents. AI is operating at 10x the speed of a human so even though it’s failing less, it’s causing more incidents per month.

I wonder if people are ready for this kind of paradox where using AI may be safer and faster but less reliable for “human time”.

▲__MatrixMan__ 21 hours ago | parent [-]

I think it comes down to just how few of our systems had a sane trajectory before the cadence sped up. If we were going somewhere, we'd have gotten there by now--aided with our new tools--but mostly we're just going in circles due to a lack of vision among leadership, so we're just going in circles faster.