AI Didn't Make Programming Easier. It Just Made It Differently Difficult
Jeremy Osborn argues in CACM that AI coding assistants have relocated the cognitive work of programming rather than reduced it. Decades of empirical research treated working memory and long-term recall as the discipline's central bottleneck; an assistant that generates boilerplate and retrieves syntax on demand lowers the penalty for imperfect recall, and Osborn reads that through distributed cognition, cognitive load theory and the extended mind hypothesis. What it does not touch is architectural reasoning, impact analysis and long-term maintenance, all of which depend on a mental model of the codebase that cannot be offloaded. The hard part moves from recall -- how do I write this -- to judgment -- does this actually make sense. Osborn draws out four consequences: the field opens to people who would have bounced off the memorisation overhead, the work becomes differently difficult rather than simpler, education bends from syntax toward systems, and the programmer survives as the orchestrating agent who decides what matters. The paradox he names is worth the read on its own: the barrier to producing code falls while the barrier to producing good code rises, because judgment is harder to develop than recall ever was.
Was this useful?