Google’s AI team unveiled two new agents aimed at easing the academic workflow. PaperVizAgent turns method sections and captions into publication‑ready figures, while ScholarPeer automates peer reviews by searching live literature and flagging missing datasets.

Both agents use a five‑stage pipeline. PaperVizAgent’s retriever, planner, stylist, visualizer, and critic iterate until the illustration matches the text, scoring 60.2 on a 0‑100 benchmark that outpaces GPT‑Image‑1.5 and Nano‑Banana‑Pro.

ScholarPeer builds a domain narrative with a historian agent and verifies claims through an adversarial scout. In public‑dataset tests, it beats state‑of‑the‑art reviewers and produces critiques that mirror senior researchers’ depth.

These prototypes demonstrate how multi‑agent orchestration can replace tedious manual steps, letting scientists focus on discovery rather than diagramming or reviewer fatigue.