Improving the academic workflow: Google Research introduces two AI agents for figures and peer review
Scientific research has always been a process that demands much more than intelligence and dedication. Anyone who has dived into this world knows that the journey between having a promising idea and seeing it published in a respected journal is long, full of bureaucratic steps, and, at many points, frustrating. It is not just about discovering something relevant. You need to communicate that discovery clearly, visually, and with technical precision, passing through rigorous quality filters before the world gets access to the final result.
Anyone who has been through the cycle of publishing a paper knows exactly what this is like: endless hours adjusting technical figures, waiting for reviewer feedback, and revising the same document for the tenth time. It is an academic workflow that wears you down, slows you down, and often frustrates even the most seasoned researchers, regardless of expertise level or institution. The feeling of going in circles is real and recurring, especially when you realize that a good chunk of your time is not going toward actual scientific reasoning but toward operational tasks that could be far more efficient.
It is no exaggeration to say that a large portion of a scientist’s time goes toward tasks that could be faster, rather than toward what truly matters — producing new knowledge. Putting together a clear, coherent, and visually appropriate figure for an international journal can take days. Waiting for a reviewer’s feedback can take months. And even after that, the cycle starts all over again with corrections, responses to reviewers, and new manuscript versions. The process is lengthy by nature, but some of that time could be reduced with the right tools. And that is where one of the most interesting developments we have seen recently at the intersection of technology and science comes in.
It was exactly with this in mind that Google Research stepped in with a pretty concrete proposal. The initiative introduces two AI agents developed specifically to tackle two of the biggest bottlenecks in the academic workflow: creating quality technical figures and the peer review process. It sounds simple in theory, but the practical impact could be much larger than it appears at first glance. 👀
What Google Research is actually proposing
Google Research published a paper describing two AI agents designed to operate directly within the scientific production process. Each agent was built with a very specific goal, and that specialization is precisely what makes them relevant. This is not a generic assistant that does a little of everything with mediocre quality. These are systems designed to understand the academic context for real, with all its quirks, conventions, and technical requirements that people outside this world rarely notice.
The first agent focuses on generating and refining scientific figures — the charts, diagrams, and visualizations that accompany any serious research paper. The second agent acts as an automated reviewer, simulating the critical eye of a specialist who evaluates the quality and consistency of a manuscript before it is formally submitted for publication. Together, they cover two distinct and equally time-consuming stages of the publishing journey for anyone doing science.
It is worth noting that these agents are not conceptual prototypes far removed from reality. They were developed using advanced large language models, trained on real academic data, and tested in scenarios that simulate the daily routines of researchers across different disciplines. This gives them a layer of sophistication that goes beyond what generic artificial intelligence tools typically offer when applied to the scientific environment.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The proposal is not to replace researchers or turn science into something automatic and soulless. The core idea is far more pragmatic than that: free up scientists’ time and energy so they can focus on reasoning, hypotheses, analysis, and data interpretation. The more mechanical and repetitive parts of the process — the ones that eat up hours but do not necessarily require an expert’s critical judgment — get accelerated with AI support. 🚀
The figures agent: why this matters so much
Anyone who has never worked in academic publishing might underestimate just how demanding the creation of figures really is. A scientific figure needs to be accurate, readable, aesthetically aligned with the target journal’s standards, and at the same time clearly communicate information that is often complex. Mistakes at this stage can result in paper rejection or correction requests that delay publication by weeks. And we are not just talking about pretty charts. We are talking about visualizations that need to meet accessibility standards, minimum resolutions, appropriate color palettes, and readability requirements that vary from journal to journal.
The Google Research figures agent was trained to understand the content of a paper, identify which data needs to be visualized, and suggest or generate graphic representations that are consistent with the scientific context presented. Beyond that, the system can iterate on those representations based on the researcher’s feedback, adjusting visual elements, scales, legends, and formatting according to each publication’s specific needs. This significantly reduces the time spent on this stage, which is often underestimated when planning any research project.
To give you a practical example, imagine a researcher finalizing a study with multiple comparative datasets. Normally, that person would need to:
- Choose the most appropriate chart type for each dataset
- Manually configure each visualization in a tool like Python, R, or even Excel
- Adjust colors, fonts, and sizes to meet the editorial guidelines of the chosen journal
- Check whether legends are correct and whether the visual information is consistent with the text
- Export in high resolution and in the correct format required by the submission platform
With the figures agent, a good portion of these steps can be automated or at least significantly sped up, without the researcher needing advanced-level skills in design or data visualization tools.
Another important point is that the agent does not just deliver a figure and call it done. It works within a conversational flow, which means the researcher can interact with the system, point out what did not work, request adjustments, and receive new versions iteratively. This is quite different from using a static visualization tool, where the user needs to master the tool itself on top of knowing what they want to communicate. With the agent, the focus stays on communicative intent, and the visual execution becomes a more natural outcome of that process.
This conversational approach also allows less experienced researchers to learn along the way. By interacting with the agent and understanding why certain visual choices are suggested, users absorb good data visualization practices naturally — almost like having a technical mentor available around the clock. It is a user experience designed to be educational without being condescending. 📊
The peer review agent: a critical eye before the critical eye
Peer review is one of the most important and, at the same time, slowest stages of the entire academic workflow. Reviewers are volunteer experts who dedicate time outside their own research schedules to evaluate other people’s manuscripts. This is valuable, but it also creates a natural bottleneck: journals receive far more submissions than available reviewers can process in a timely manner. The result is wait times that can stretch for months, and sometimes over a year. It is an equation that simply does not add up when you look at the growing volume of global scientific output.
Google Research’s second agent tackles this problem from a smart angle. Instead of trying to replace human reviewers, it functions as a preliminary step — a kind of automated pre-review that identifies structural problems, methodological inconsistencies, argumentative gaps, and clarity issues before the paper even reaches a flesh-and-blood reviewer. This means that when the manuscript finally lands on a human reviewer’s desk, it has already been through a rigorous filter that increases the chances of the work being more mature and robust. The average quality of submissions tends to go up, and the process as a whole becomes more efficient for everyone involved.
This positioning as a complementary step, rather than a replacement, is critical for the tool’s acceptance by the scientific community. Researchers and editors have legitimate reservations about delegating peer review entirely to automated systems, and Google Research seems to have understood this dynamic well by positioning the agent as an assistant that prepares the ground rather than making final decisions about the scientific validity of a piece of work.
From the researcher’s perspective, this agent works as a technical conversation partner available at any time, without the social awkwardness of asking a colleague to review your work at 11 PM on a Friday. The system reads the manuscript, evaluates it based on established academic criteria, and returns a detailed diagnosis with areas of concern and improvement suggestions. It is an extra layer of quality that previously depended entirely on your network of contacts and the goodwill of other researchers, and can now be accessed independently and repeatedly throughout the writing process. 🧠
Practical benefits that go beyond efficiency
Beyond saving time, the review agent brings a benefit that is easy to overlook but hard to overstate: it reduces the anxiety of the submission process. Every researcher knows the knot in their stomach when clicking the submit button without being sure the work is truly ready. With an automated pre-review, that insecurity decreases because the author already has a clearer picture of where the manuscript’s weak points are and can fix them before being exposed to the formal judgment of external reviewers.
Another interesting aspect is the potential to level the playing field in submission quality. Early-career researchers, for example, often have more difficulty identifying structural problems in their manuscripts simply because they are still developing that critical eye. The review agent can function as a leveling tool, giving those researchers feedback that would normally only come after years of accumulated experience in the publication process.
The question of bias and transparency
Not everything is rosy, of course. The introduction of AI agents into the scientific process raises legitimate concerns that need to be discussed seriously. One of the most relevant is the question of bias. Language models are trained on existing data, and that data reflects decisions, priorities, and inevitably, biases that already exist within the scientific community. If the review agent learned to evaluate manuscripts based on papers published in specific journals, there is a risk that it reproduces preferences for certain types of methodology, writing formats, or even research fields.
Google Research has not yet publicly detailed all the bias mitigation mechanisms implemented in these agents, but the awareness that this is a real problem is already an important step. The scientific community will certainly push for transparency on this front, and it is likely that independent audits and continuous model reviews will be part of the maturation process for these tools over time.
Another important issue is the privacy of scientific data. Researchers who submit manuscripts to AI systems before publication need clear guarantees that their data and discoveries will not be leaked, used to train other models without consent, or somehow compromised before official publication. This is a sensitive point that could determine the level of adoption these tools see from the academic community.
What this means for the future of science
Google Research’s move with these two AI agents is not an isolated event. It is part of a broader movement integrating artificial intelligence and scientific production that has been gaining momentum in recent years. AI tools are already being used for literature reviews, data synthesis, pattern identification in large databases, and even for suggesting new research hypotheses. What Google did here was identify two specific friction points in the academic workflow and build targeted solutions for them — an approach that is far more effective than creating a generic system and hoping it turns out useful somehow.
For the scientific community, this raises interesting questions about the role of AI as a partner in the knowledge-building process. Automated peer review, for example, is already sparking debates about how to ensure the AI’s perspective does not reproduce biases present in the training data, or how to guarantee that automated evaluations are truly independent from the interests of whoever developed the system. These are legitimate questions that need answers over time, and they will likely shape how these tools are regulated and adopted by major scientific journals.
The picture taking shape is one of science that is more assisted by technology but still fundamentally depends on human judgment for the decisions that truly matter. Google Research’s agents do not decide whether a hypothesis is valid. They do not determine whether a result is significant. What they do is handle the operational side with competence, allowing the researcher to dedicate more time and energy to the questions that require creativity, intuition, and critical thinking.
Even so, it is hard to ignore the concrete potential these initiatives have for democratizing access to quality research. Researchers at institutions with fewer resources, in countries with less academic support infrastructure, or in fields with few specialists available for review gain, with tools like these, a level of support that was previously reserved for those with access to large international collaboration networks. That is no small thing. It is a change that could, over time, genuinely alter who gets to publish quality science and under what conditions that happens. 🌍
Whether these agents will become standard in daily academic life is still too early to call. But the signal Google Research is sending is clear: artificial intelligence has the potential to make the scientific process faster, more accessible, and more equitable — as long as it is used as a support tool and not as a substitute for the human thinking that drives science forward.
