Superhuman is redefining email by embedding AI deeply into workflows without sacrificing speed, ensuring every feature—from auto-labeling to 'Ask AI'—adds value without latency. The team combats 'agent laziness' through rigorous evaluation frameworks, including custom tests like date reasoning and semantic search across massive inboxes. They prioritize local-first architecture for reliability and speed, using on-device storage and exploring offline AI models. Cost is managed via box-priced inference and long-term controllability over per-token models. AI isn't just a tool—it's evolving into a proactive executive assistant capable of drafting, scheduling, and understanding personal context. Internally, AI has boosted engineering output from 4 to 6 PRs per week, highlighting a cultural shift toward tool empowerment. Yet, as productivity rises, questions emerge about compensation and fairness. Loïc stresses that strong coding fundamentals remain essential, as AI amplifies skilled engineers while exposing those without deep understanding.