AI Agents Move From Chatbots to Autonomous Scientific Discovery
By @sharedot · · 8 pages
- Tech
- AI Agents
- Scientific Discovery
- Autonomous AI
- Biomedical Research
Three landmark AI agent systems reported in Nature in 2026 signal a shift from passive AI assistance to active participation in scientific.
A New Class of AI Agent Arrives
Professor Guang-Guo Ying of South China Normal University examines three autonomous systems reported in Nature in 2026—ERA, The AI Scientist, and MIRA. These systems do far more than summarize papers or analyze datasets on instruction.
They can formulate or refine tasks, use software tools, generate scientific outputs, evaluate results, and take actions within defined environments, marking a major shift from passive assistance toward active participation in research.
Ying's article was published in Artificial Intelligence & Environment with DOI 10.66178/aie-0026-0017. According to bioengineer.org, the news article about AI scientists was published on August 20, 2026, with the underlying research article's news publication date listed as August 13, 2026.
What Each System Can Do
ERA is designed to develop and improve scientific software across fields including single-cell analysis and epidemiological forecasting, inspecting existing code, identifying weaknesses, and iteratively improving programs using computational evidence.
The AI Scientist can generate research ideas, write code, conduct experiments, prepare figures, draft manuscripts, and perform automated peer review—creating an end-to-end chain that resembles a full scientific investigation.
MIRA brings autonomy into medicine: it can interact with simulated electronic health records, order tests, generate diagnoses, prescribe treatments, and recommend hospital admission, representing a transition from medical question answering to sequential decision-making in clinical environments.
How AI Agents Differ From Chatbots
AI Agents are computational systems that can autonomously execute workflows using a variety of tools and resources.
Milton Pividori, assistant professor of biomedical informatics at CU Anschutz, explains that a chatbot is just an interface to a system that could be a single agent or a multi-agent system.
The architecture of a multi-agent system usually has an orchestrator agent that coordinates tasks, delegating to specialized sub-agents—a literature review agent, a hypothesis generation agent, or a Python experiment agent.
This task isolation matters because LLMs have limited context: as context grows, the model gets confused and hallucinates more.
A recent publication in Nature introduced a tool called Robin that can review literature, develop hypotheses, suggest experiments and analyze data, enabling what its authors called a semi-autonomous approach to scientific discovery.
Growing Interest and Real-World Lab Use
Public interest in AI Agents has risen more than 550% from the previous year according to Google Trends.
Yet AI agents have been developed since almost two decades ago, as Yanjun Gao, assistant professor of biomedical informatics and co-director of the Center for Health AI (CHAI) at CU Anschutz, illuminates—earlier systems used probabilistic language models and slot-filling NLP to book airline tickets or train tickets.
The Pividori Lab received a grant to support research on Type 1 diabetes using AI Agents, leveraging multi-agent systems to integrate knowledge bases and tools to generate hypotheses about gene modules and their interactions.
The goal is human-AI interaction where the human provides key insight and the AI augments with mechanical, time-consuming work.
Benchmarking Discovery Itself
As AI moves deeper into science, the distinction between knowing and discovering becomes critical. According to PR Newswire, Apodex introduced TRACES, a benchmark designed to evaluate AI on real-world scientific discovery problems where the answer may not yet be known.
According to PR Newswire, TRACES assembled 423 high-value problems from a survey of 561 industries across 16 sectors. Rather than scoring only a final answer, TRACES gives AI systems executable environments with scientific literature, datasets, code execution, and experimental feedback.
According to PR Newswire, Apodex is headquartered in Redwood City, California. According to PR Newswire, the TRACES technical paper is available on arXiv.
According to PR Newswire, Apodex launched TRACES on August 19, 2026. According to PR Newswire, Dr. Sheng Wang is Lead Scientist at Apodex leading discovery and evaluation, and Brian Wang is AI Research Scientist at Apodex.
According to PR Newswire, Apodex's founding thesis is that scale and data alone cannot cross the barrier from pattern-matching to genuine discovery.
Stakes, Limits, and the Path Forward
Autonomous AI systems can generate hallucinated citations, incorrect code, duplicated figures, and physically impossible results, and their outputs can be difficult to reproduce when prompts, software versions, and intermediate decisions are not adequately recorded.
Ying argues that human scientists will remain essential for defining important questions, setting ethical boundaries, and evaluating ambiguous findings. The systems' growing autonomy raises a central issue: how responsibility is assigned when an autonomous system makes a chain of decisions.
MIRA operates in simulated electronic health records, and its reported capabilities should not be confused with unrestricted clinical independence—real-world medicine contains ambiguity difficult to reproduce in simulation.
The decisive question is no longer whether machines can participate in scientific work, but whether humans can build the institutions and safeguards needed to ensure that autonomous research expands knowledge without weakening scientific standards.
Sources
- bioengineer.org › AI Scientists Are Reshaping Scientific Discovery Through Autonomous Research Agents
- miragenews.com › AI Agents Revolutionize Scientific Discovery
- news.cuanschutz.edu › AI Agents in Science: What Are AI Agents, and How Are They Being Used in Scientific Research?
- prnewswire.com › Apodex Launches TRACES, a Benchmark for Scientific Discovery