AI Agents in Scientific Research

An overview of how agentic AI systems are being used to automate complex scientific workflows. These systems can solve competitive programming problems, autonomously generate hypotheses and execute lab experiments, and perform chemistry tasks like synthesis planning. The approach involves integrating reasoning, experimentation, and learning into repeatable processes, moving from single-purpose tools to fully integrated systems for research.

AI Agents in Research and Administration

Gemini 2.5 Deep Think at ICPC Gemini 2.5 Deep Think secured a gold medal at the 2025 International Collegiate Programming Contest World Finals. The system solved 10 out of 12 problems, placing it second overall compared to human teams. It resolved one problem that no human competitor could. This performance builds on its prior success at the International Mathematical Olympiad.
Robot Scientist Adam The autonomous system Robot Scientist Adam generated hypotheses, designed and executed experiments using lab automation, and interpreted results in yeast functional genomics.
ChemCrow Chemistry Agent ChemCrow, an LLM-based chemistry agent, iteratively plans actions and utilizes expert chemistry tools and databases. It addresses tasks such as synthesis planning, drug discovery, and materials design.
AILA for Atomic Force Microscopy AILA is an LLM-agent framework that automates atomic force microscopy (AFM) workflows. It handles planning, calibration, data capture, and analysis, with its performance assessed using the AFMBench benchmark suite.
Autonomous Scientific Workflow Orchestration AI agents are enabling autonomous scientific workflow orchestration, shifting research from isolated tools to integrated systems. These agents plan, execute, evaluate, and iterate through research pipelines, connecting reasoning, experimentation, and learning into repeatable processes.
Hermes Agent for Product Research An open-source example demonstrated AI agents for product research, using the Hermes Agent integrated with NVIDIA NemoClaw for data processing and NVIDIA OpenShell for security. This system functions across platforms like Outlook, Slack, and GitHub.

Sources

Gemini 2.5 Deep Think achieves gold-medal level at ICPC 2025 - Notebookcheck News
The Gemini 2.5 Deep Think AI achieved gold-medal level at the 2025 International Collegiate Programming Contest (ICPC). It solved ten of twelve tasks, including one that none of the participating human teams could solve. This achievement demonstrates progress in abstract problem solving, programming and the potential for AI-based developer support.
Gemini achieves gold-medal level at the International Collegiate ...
Gemini 2.5 Deep Think has achieved gold-medal level performance at the 2025 International Collegiate Programming Contest (ICPC) World Finals, building on a previous gold-medal win at the International Mathematical Olympiad (IMO). This demonstrates a significant leap in abstract problem-solving, with Gemini solving 10 out of 12 problems and achieving a 2nd place overall ranking if compared with university teams. The success highlights AI's potential as a problem-solving partner for programmers and its ability to contribute unique solutions to complex challenges.
Gemini wins International Collegiate Programming ... - Google Blog
Gemini achieved gold-medal performance at the International Collegiate Programming Contest World Finals. An advanced version of Gemini 2.5 Deep Think achieved gold-medal level performance at the 2025 ICPC World Finals, building on a previous win at the International Mathematical Olympiad. Gemini solved Problem C, a complex optimization task that no university team could solve, demonstrating its world-class abstract problem-solving capabilities.
Gemini 2.5 Deep Think scores competitive coding gold ... - 9to5Google
Gemini 2.5 Deep Think achieved a gold-medal level performance in competitive coding at the International Collegiate Programming Contest (ICPC), solving 10 out of 12 problems in 677 minutes. This advanced version of Gemini 2.5 Deep Think would have ranked second overall compared to human teams. Google highlights Gemini's ability to solve a particularly difficult problem (Problem C) that no human teams could solve, demonstrating a "profound leap in abstract problem-solving."
medium.com
AI agents for scientific workflow automation: From hypothesis to experiment | by Khayyam H. | Medium After years spent moving between research labs, classrooms, and production engineering teams, one truth has stayed uncomfortably consistent: modern science is computationally advanced but operationally fragmented. We run large-scale simulations, train sophisticated models, and process massive datasets, yet the connective tissue between thinking, testing, and learning remains overwhelmingly manual. In practice, hypothesis generation, experiment design, execution, and analysis are still…
Deploy Self-Evolving Agents for Faster, More Secure Research with a Hermes Agent and NVIDIA NemoClaw
AI agents are powerful for synthesizing data to accelerate research, but combining internal with public data presents security challenges. This post shares an open-source example using Hermes Agent with NVIDIA NemoClaw for product research across Outlook, Slack, and GitHub, with NVIDIA OpenShell enforcing a secure runtime. The agent learns user preferences and patterns, improving with increased interaction.
Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad
chemistryworld.com
Co-Scientist: A multi-agent AI partner to accelerate research ...
Accelerating scientific breakthroughs with an AI co-scientist

Full transcript

AI is now being developed to automate entire scientific workflows, from hypothesis to experiment. This is Research Agent, and today we are examining how these autonomous systems are being applied to discovery. Agentic AI systems are being developed to autonomously orchestrate complex scientific workflows. These agents can plan, execute, evaluate, and iterate through research pipelines, connecting reasoning with experimentation and learning. One such system, Gemini 2.5 Deep Think, has demonstrated its capability in competitive problem-solving. At the 2025 International Collegiate Programming Contest World Finals, it solved 10 of 12 problems. According to the DeepMind and Google blogs, this performance would have ranked it second among human university teams. The system previously achieved a gold medal at the International Mathematical Olympiad. In other scientific domains, agents are managing physical experiments. The Robot Scientist Adam autonomously generated hypotheses and conducted lab work in yeast functional genomics. In chemistry, an agent named ChemCrow handles tasks like synthesis planning and drug discovery by combining a large language model with specialized tools. Other frameworks, like AILA, automate entire workflows for scientific instruments such as atomic force microscopes, from planning to data analysis. This approach signifies a move from using single-purpose AI tools to deploying more integrated systems for research. We'll have more on the application of AI to scientific research next time. Thanks for listening to Research Agent.