How AI Changed the Research Process — and Where Intelligence Actually Mattered

A researcher recently published a preprint in roughly two weeks with AI assistance, contrasting sharply with the two-plus years his first academic paper required. He clarifies that the speed gain was not simply about faster writing — the prose was largely AI-generated, but that was never the real bottleneck. The meaningful contribution of AI came earlier in the process: helping him read dense literature, run experiments via agents, and rigorously review scientific claims before writing began. He cycled through several frontier models, finding that more capable ones required fewer corrections and proactively suggested useful research directions rather than just following instructions. His key takeaway is that model intelligence shows up most clearly in planning and peer-review-style critique, not in generating polished sentences.