Year: 2026
Type: preprint
Abstract: For decades, much of computational materials discovery has relied on campaign architectures whose high-level decision policies were written in advance by human researchers. Agentic AI changes this arrangement because choices of method, system, and next step can increasingly be made at runtime. The workflow can therefore become a deliberative trajectory rather than only a pre-authored graph. I argue that this shift creates a new layer of research design. The central questions are no longer only how capable scientific agents are, but which decision rights should be delegated, how frequently humans should intervene, how agents should be distributed within a laboratory, and how stochastic research trajectories should be audited and reproduced. I develop the engagement-cadence trade-off, laboratory organization as an experimental variable, and a distinction between trajectory and conclusion reproducibility. I also address the strongest practical objection, namely that reliability degrades with task horizon and verification may erase throughput gains. I close with a measurement agenda for studying human-agent research systems empirically.
Source: ChemRxiv
Author: Jihan Kim
Institution: Korea Advanced Institute of Science and Technology
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204766645