Benchtalks #3: We taught AI everything except how to learn

An interview with researcher Parth Asawa on why agents re-read an entire codebase every session rather than building persistent understanding, and how benchmarks can measure real learning ability.