New Anthropic research surveyed 1,260 quantitative social scientists on AI use, and the headline finding is a gap I find hard to ignore: 81% have tried AI chatbots, but only 20% use coding agents like Claude Code that actually write and run analysis autonomously.

What struck me, coming from international politics and history rather than a stats-heavy background, is how uneven adoption already is. Economists are at 39%, political scientists 25%, but communications, education and public health sit in the single digits.

Men adopt at more than twice the rate of women. Researchers at top universities are 40% more likely to use these tools. So the disciplines and people best placed to study inequality are themselves splitting along familiar lines in who gets to use the most powerful new method.

Coding agent users post more working papers and chase more grants, but they aren’t yet submitting more to journals. And researchers are noticeably more optimistic about AI helping them write papers than about whether it’s good for their field overall.

As someone trained to think about how tools shape the questions we ask, that last gap feels like the real story: when machines start making the analytical choices, they quietly shape how we come to understand our economy and our politics.