Some international team of researchers spent artificial intelligence system capable of independently conducting research in multiple disciplines – generating articles from initial concept to publication-ready manuscript in about 30 minutes at about $4 each.
The so-called system denariuscan formulate research ideas, review existing literature, develop methodologies, write and execute code, create visualizations and prepare complete academic papers. In a demonstration of their versatility, the band used Denario to generate articles spanning astrophysics, biology, chemistry, medicine, neuroscience and other fields, with one AI-generated paper already accepted for publication in scientific conference.
“Denario’s goal is not to automate science, but to develop a research assistant that can accelerate scientific discovery,” the researchers wrote in a paper published Monday describing the system. The team creates software publicly available as an open source tool.
This achievement represents a turning point in the utilize of immense language models in scientific work, potentially changing how researchers approach early-stage research and literature reviews. However, the study also highlights significant limitations and raises pressing questions about validation, authorship, and the changing nature of scientific work.
From data to draft: How AI agents collaborate to conduct research
At its core lies denarius it operates not as a single AI brain, but as a digital research department where specialized AI agents collaborate to take a project from concept to completion. You can start the process with “Idea module” which uses a fascinating adversarial process in which “Idea Creator“the agent proposes research projects that are then analyzed by”Hater of ideas” that critiques them on feasibility and scientific merit. This iterative loop refines raw concepts into solid lines of inquiry.
Once the hypothesis is confirmed, “Literature module” searches academic databases such as Semantic Scholar to check the novelty of an idea, then follows “Methodological Module” which presents a detailed step-by-step research plan. The heavy lifting is then done by “Analytical module“, a virtual workhorse that writes, debugs, and executes its own Python code to analyze data, generate graphs, and summarize results. Finally, “Paper module” downloads the obtained data and prepares graphs, and then prepares a complete scientific paper in LaTeX, which is a standard in many fields of science. In the final, recursive step “Review module” can even act as an AI reviewer, providing a critical report on the strengths and weaknesses of the generated work.
This modular design allows the researcher to intervene at any stage, presenting their own idea or methodology, or simply using Denario as an end-to-end autonomous system. “The system has a modular architecture, so it can handle specific tasks, such as generating an idea or conducting comprehensive scientific analyses,” the article explains.
To verify its capabilities, the Denario team put the system to the test, creating a huge repository of articles in many fields. As a striking proof of concept, one article generated entirely by Denario was accepted for publication on the site Agents4Science 2025 conference — a peer-reviewed place where the main authors are the artificial intelligence systems themselves. The paper, titled “QITT-Enhanced Multi-Scale Substructure Analysis with Learned Topological Embeddings for Cosmological Parameter Estimation from Dark Matter Halo Merger Trees,” successfully combined complex concepts from quantum physics, machine learning, and cosmology to analyze simulation data.
Ghost in the machine: ‘nonsense’ artificial intelligence results and ethical alarms
While the successes are noteworthy, the research paper is refreshingly honest about Denario’s significant limitations and failure modes. The authors emphasize that the system currently “behaves more like a good undergraduate or early graduate student than like a full professor, in terms of the substantial picture, combining results…etc.” This honesty provides a crucial reality check in a field often dominated by noise.
The article devotes entire chapters to “Failure modes“And”Ethical implications“, a level of transparency that enterprise leaders should pay attention to. The authors report that in one case, the system “hallucinate an entire article without implementing the necessary numerical solution,” inventing the results to fit a plausible narrative. In another test for a purely math problem, the AI created text that had form mathematical proof, but according to the authors it was “mathematically empty”.
These failures highlight a critical point for any organization looking to implement agent-based AI: Systems can be brittle and prone to reliable-sounding errors that require expert human supervision. Denario’s article serves as an important case study in the importance of informing people about the need for verification and critical evaluation.
The authors also face the deep ethical issues raised by their work. They warn that “AI agents could be used to quickly flood the scientific literature with claims driven by a particular political agenda or specific commercial or economic interests.” They also touch on the “Turing trap” – a phenomenon that aims to mimic human intelligence rather than enhance it, potentially leading to a “homogenization” of research that stifles real, paradigm-changing innovation.
An open source second pilot for labs around the world
Denario is not just a theoretical exercise locked in an academic laboratory. The whole system is there open-source under the GPL-3.0 license and is available to the wider community. The main project and its graphical user interface is DenarioApp available on GitHubwith installation managed using standard Python tools. For enterprise environments focused on reproducibility and scalability, the project also provides official Docker images. Public demonstration organized on Hugging facial space allows everyone to experiment with its capabilities.
For now, Denario remains what its creators call a powerful assistant, but it cannot replace the keen intuition of a human expert. This framing is intentional. Denario’s project is less about creating a robotic scientist and more about building the ultimate co-pilot, designed to handle the tedious and time-consuming aspects of newfangled research.
By handing over the grueling work of coding, debugging, and initial drafting to an AI agent, the system promises to free researchers from the one task it cannot automate: the deep, critical thinking required to ask the right questions.