1 episode
If context windows jump to 10M tokens, which AI engineering practices survive? We separate load-bearing architecture from temporary scaffolding.
