Google’s New AI Agents Want to Draw Your Figures and Review Your Papers

Google Research just dropped two new AI agents aimed at the academic workflow, and honestly, they address pain points I’ve been complaining about for years.

One is PaperVizAgent (formerly called PaperBanana, which I kind of preferred), an agent that generates publication-ready figures from your manuscript text. The other is ScholarPeer, an automated reviewer that evaluates papers, including inline diagrams.

Let me break down what these actually do and whether they’re worth your time.

PaperVizAgent: Finally, someone to draw the diagrams I keep putting off

If you’ve ever submitted to a top-tier conference, you know the drill: you spend weeks perfecting the text, then realize you need methodology diagrams, statistical plots, and god knows what else. Creating these from scratch is a slog.

PaperVizAgent takes two inputs: your source context (typically the method section) and a figure caption describing what you want. Then it orchestrates five specialized sub-agents:

• A retriever that grabs relevant references from existing literature
• A planner that organizes the content
• A stylist that enforces academic aesthetics (no Comic Sans, thank you)
• A visualizer that renders the image or generates Python code for plots
• A critic that checks the output against the original text and triggers refinements if something’s off

The iterative loop is key. The critic catches inconsistencies and feeds back to the visualizer, so the final figure isn’t just pretty — it’s technically accurate.

From what I’ve seen in their examples, the results are genuinely impressive. They claim PaperVizAgent consistently outperforms GPT-Image-1.5 and other baselines. I’d believe it. The sample methodology diagrams look like something you’d see in a NeurIPS proceedings.

ScholarPeer: An automated reviewer that actually reads the paper

Peer review is broken. The volume of submissions has exploded, reviewers are exhausted, and evaluations are all over the place. ScholarPeer tries to fix this by acting as a rigorous automated reviewer.

It doesn’t just skim abstracts. It reads the full paper, including diagrams, and produces structured feedback covering methodology, results, and literature grounding. The agent checks claims against cited sources and flags missing references.

Their evaluation shows ScholarPeer beats existing automated reviewers by a solid margin. I’m skeptical about replacing human reviewers entirely — there’s nuance that AI still misses — but as a first-pass filter or consistency check, this could save editors and program committees a lot of headache.

The risk, of course, is that authors start optimizing for the AI reviewer instead of for scientific quality. But that’s a problem with any automated system.

What this means for researchers

These tools are still early, but the direction is clear. AI is moving from being the subject of academic papers to being an active participant in producing them.

PaperVizAgent could save hours per figure. ScholarPeer could make the review process faster and more consistent. Neither is perfect, but they’re both useful right now.

I’d love to see the code released for PaperVizAgent — they’ve already shared a paper and code link. ScholarPeer is still in paper stage.

If you’re tired of wrestling with Illustrator or getting inconsistent reviewer feedback, keep an eye on these. They might be the productivity boost your lab needs.