all these focus on long running agents without focussing on core restructure is baffling. the immediate need is to break down complex tasks into smaller ones and single shot them with some amount of parallelism. imo - we need an opinionated system but with human in the middle and then think about dreamy next steps. we need to focus on groundedness first instead of worrying about agent conjuring something from thin air. the decision to leap frog into automated long running agents is quite baffling.
boys are trying to single shot a browser when a moderate complex task can derail a repo. there’s no good amount of info which might be deliberate but from what i can pick, their value add was “distributed computing and organisational design” but that too they simplified. i agree that simplicity is always the first option but flat filesystem structure without standards will not work. period.