The Agentic AI Summit 2025 at UC Berkeley brought together leading researchers to explore how autonomous AI systems are reshaping scientific discovery. Speakers emphasized these tools’ ability to plan multi-step experiments and adapt strategies based on real-time data – a key advantage for fields like drug development. IBM’s latest models now feature improved chain-of-thought training, allowing AI agents to explain their reasoning process step-by-step when analyzing complex problems like protein folding.

Forrester’s new industry analysis reveals that 73% of major research institutions plan to adopt agentic AI within two years. These systems excel at orchestrating lab equipment and cross-referencing findings across chemistry, biology, and clinical trial data – tasks that previously required specialized teams.

In practical applications, Causaly demonstrated how their hypothesis-generating AI identified three promising biomarkers for rare diseases by analyzing over 10 million medical studies. The system autonomously mapped connections between genetic data and clinical outcomes, proposing viable Phase II trial candidates that human researchers had overlooked.

Global collaboration efforts are emerging as labs in Germany and Japan begin sharing AI agent protocols for materials science. Early results show machine learning models can now predict chemical reaction outcomes with 89% accuracy – up from 62% in 2024 – by continuously learning from failed experiments.

Ethical discussions dominated the Berkeley summit’s closing sessions, with panels establishing guidelines for AI transparency in research. New standards require agentic systems to maintain detailed “decision logs” showing how they prioritize tasks and validate findings against peer-reviewed studies.