It’s certainly #2. They have shown over dozens of decisions they move very quickly, break stuff, then have to both figure out what broke and how to explain it.

It’s definitely a cost / resource saving strategy on their end.

It's very weird that they frame caching as "latency reduction" when it comes to a cloud service. I mean, yes, technically it reduces latency, but more importantly it reduces cost. Sometimes it's more than 80% of the total cost.

I'm sure most companies and customers will consider compromising quality for 80% cost reduction. If they just be honest they'll be fine.

The same company that claims they have models that are too "dangerous" to release btw.

what's even more amazing is it took them two weeks to fix what must have been a pretty obvious bug, especially given who they are and what they are selling.

they just vibecoded a fix and didnt think about the tradeoff they were making and their always yes-man of a model just went with it