The prevailing wisdom right now is that you should put all your code in a monorepo so that AI has all the context to act. I don’t think so. As context windows get bigger, effectiveness of AI drops off (context rot) and token usage gets higher. Instead, the more tightly scoped you can keep the problem space the better. So what I think this means for AI native system design is that microservices or lambdas are the best play.
The issues with microservices don’t go away though. You need to coordinate multiple services, and stay on top of contracts and system complexity. So the challenge is to build out a system that handles that architecture layer, so that the smaller agents can handle the systems, and the whole thing functions like it’s supposed to.
“The weight in this set increased by 3,339.17% over the last time you logged this exercise.”
Think my fitness app must reduce the amount of weight it expects when not in use, and I haven’t been to the gym in over a year 😬
This was 4kg
The problem with the new F1 regs isn’t the super clipping or the new boost.
It’s that the viewer can’t understand the strategic position of the race because we don’t know who has got what battery level remaining. So drivers just seem to speed up or slow down at random.
@pelaseyed@Jason Yeah I’m convinced that most of the AI layoffs we’re seeing are a case of “we have to lay people off but if we say it’s AI then we can still have the market think this is a good news story”
@af3@gunnarmorling Exactly! The dollars spent of one person vs another is not what you’re looking at. Instead you’re looking for a signal of “are these people trying to use the latest tools to speed themselves up or not”
@cricket_badger@keeps71 Reckon this should apply only if the bowling team has managed 85 overs in the day. Would give them a good incentive to speed things up a bit