A PNAS reanalysis shows a headline result on LLM-driven research productivity can be reproduced with random, inverted and pre-ChatGPT placebo treatments — plus AI agents that can't yet reproduce their own field's papers, platform AI policies that drive creators away, and LLM-simulated bank runs.

AI agents are fact-checking the published canon and turning up decades-old errors in papers and reference tables — while journals debate whether AI should be patching peer review or replacing it. Plus: the next Turing tests, generative AI as a causal tool, and a model predicting we'll 'do more, less well'.

A Strategic Management Journal framework for running experiments with LLM-powered agents, a PNAS essay on whether research stays a human enterprise, and store cameras turned into an 'offline clickstream'.

A 7.3-million-article study says most published papers now show signs of LLM influence — and a new PNAS exchange argues the yardstick itself is broken. Plus: when AI teams don't need to coordinate, ML from 14 examples, and 500 years of art in latent space.