It is argued that recent LLM-based agent harnesses enable an emerging form of scientific software, agentic programs, that combine deterministic algorithms with bounded LLM-based judgment, task-specific verification, episodic maturation, and complete delegation in production.
Abstract
Computational materials science has traditionally delegated algorithmic tasks to computers while leaving scientific judgments to humans. We argue that recent LLM-based agent harnesses enable an emerging form of scientific software, agentic programs, that combine deterministic algorithms with bounded LLM-based judgment, task-specific verification, episodic maturation, and complete delegation in production. We illustrate this concept with DeMARS, an agentic program for constructing atomistic models from experimentally measured disordered crystal structures.
By connecting the heterogeneous stages of computational materials discovery, the LLM-based agents of MAESTRO can operate across application domains and uncover high-performance materials that conventional screening approaches would be unlikely to consider.
Yun-Tong Chen, Ju Huang, Yu Liu et al.· 0 citations
Scientific software presents a demanding test for computer-using agents based on visual language models (VLMs): completing a research workflow requires interpreting specialized interfaces, manipulating scientific objects, and producing verifiable results. We thus introduce OSWorld-Science, a benchmark and evaluation en...
Ding-Yuan Dai, He-Li Qi, Lei Liu et al.· 0 citations
The speed and scale of disruption agentic systems are bringing to computational chemistry leave many of us dumbfounded about the field's future and what the authors should spend their efforts on, as already established specialists, teachers, and students, and they have no answer.
Pavlo O. Dral, Hassan Nawaz, Arif Ullah· 0 citations
Scientific software is often locked in its source language, callable from others but not natively extensible or built upon. We present Bio-Babel (biobabel.stanford.edu), an autonomous framework that rebuilds software natively in target ecosystems and makes it agent-callable, shipping each package with an agent-readable...
Nianping Liu, Xuan-Zhi Chen, Miao Cui et al.· bioRxiv· 0 citations
As machine intelligence becomes part of science's epistemic infrastructure, steps must be taken to ensure that researchers who did not build, own, or operate the system can still contest how ideas and evidence became a claim.
CompMat-Bench is introduced, a benchmark of 94 tasks derived from recently published computational materials studies, each asking agents to complete a step toward achieving the study's scientific goal, and provides a basis for comparing agents on the steps of real materials research and for analyzing agent failure mode...
Chen-Mu Zhang, Levi Felix, Jun-Jie Zhang et al.· 0 citations
