Everyone now knows that LLMs can’t make good decisions about complex software design at scale. None moreso than everyone working at and invested in frontier labs. So far, no conventional wisdom about what LLMs can’t do has survived impact. We will likely find out within 8 months whether we are entering a new AI winter or we passed the event horizon where capital will fully replace labor.

Personally, I don’t see how it can be done. But I can’t trust my judgement. It has never survived impact.