Google’s new AI agents want to fix your academic figures and peer review

Google Research just dropped two new AI agents aimed at the academic workflow, and honestly, they might actually be useful. One draws figures for you, the other reviews papers. Let’s talk about both.

PaperVizAgent: Finally, figures that don’t look like garbage

If you’ve ever spent a weekend wrestling with Adobe Illustrator or matplotlib trying to make a methodology diagram that doesn’t look like a toddler’s scribble, you’ll appreciate PaperVizAgent. It’s an autonomous framework that takes your manuscript text and a figure caption, then generates publication-ready illustrations.

The system uses five specialized agents: a retriever, a planner, a stylist, a visualizer, and a critic. The retriever and planner grab references from existing literature and organize the content. The stylist ensures the output matches academic standards—no Comic Sans or neon colors. The visualizer renders the image or generates executable Python code for statistical plots. Then the critic checks everything against the original text and kicks back feedback if something’s off. It’s an iterative refinement loop, and it actually produces figures that look like they belong in a top-tier conference.

I’ve seen a lot of “AI for figures” tools, and most of them are garbage. PaperVizAgent is different. It consistently beats baselines like GPT-Image-1.5 and Nano-Banana-Pro. The examples in their paper show methodology diagrams that are actually clear and precise. I’m not saying it’ll replace a skilled graphic designer, but for most researchers, it’s probably good enough.

ScholarPeer: The reviewer that doesn’t get tired

Peer review is a mess. The number of paper submissions has exploded, and reviewers are burned out. ScholarPeer is an AI agent designed to automatically and rigorously evaluate academic papers, including inline diagrams. It doesn’t just give a thumbs up or down—it delivers critical, literature-grounded reviews.

The key here is that ScholarPeer actually looks at the figures, not just the text. That’s a big deal. A lot of automated reviewers just scan the abstract and make a guess. ScholarPeer claims to beat state-of-the-art automated reviewers in terms of critique quality and groundedness. I haven’t tested it myself, but the reported results are impressive.

Why this matters

The academic research workflow is brutal. You spend months on a project, then more months writing, then more months dealing with reviewers who might not even read your paper carefully. Tools like PaperVizAgent and ScholarPeer could reduce the administrative overhead and let researchers focus on actual science.

Of course, there are caveats. AI-generated figures might still miss nuances that a human designer would catch. And automated peer review could introduce biases or miss subtle flaws. But Google has open-sourced the code for PaperVizAgent, so the community can poke at it and improve it.

The bottom line

Google is betting that AI can be more than just a subject of study—it can be an active participant in the scientific process. PaperVizAgent and ScholarPeer are steps in that direction. They’re not perfect, but they’re a hell of a lot better than what we have now. If you’re a researcher drowning in figure-making or review requests, these might be worth a look.