Page 132 - Read Online

Page 24 of 31 Li et al. J. Mater. Inf. 2026, 6, 10

collaboration, ultimately moving toward an advanced scientific research infrastructure capable of adapting to
complex real-world scenarios. These advancements in automated materials discovery jointly drive the rapid
development of agent-driven SDLs toward greater generality, intelligence, and environmental adaptability,
bringing revolutionary acceleration to materials creation.

CONCLUSIONS AND OUTLOOK
AI agents are reshaping the research paradigm of materials development with unprecedented depth and
breadth. From traditional trial-and-error approaches based on experience, to data-driven high-throughput
screening, and now to “Agentic Material Science” based on LLMs with autonomous reasoning and
decision-making capabilities, materials research is advancing towards a new era of high intelligence and
automation. By deeply integrating natural language understanding, multimodal knowledge integration, and
tool usage capabilities, agents have successfully connected the entire innovation chain from material design
to synthesis and characterization. This not only significantly enhances the efficiency and breadth of material
exploration, but also promotes a fundamental shift from experience-driven research to agent-driven
autonomy.

Looking ahead, Agentic Material Science will further develop around three key directions to unlock greater
innovation potential. In materials design, the integration of multimodal, cross-scale collaborative approaches
will enable AI agents to process and fuse data from different modalities and scales, thereby generating new
material designs that transcend traditional disciplinary boundaries. In synthesis and characterization, the
integration of embodied intelligence equipped with real-time perception and adaptive control will enable AI
agents to perceive, reason, and act within physical laboratory environments, allowing them to dynamically
adjust synthesis schemes based on real-time feedback from characterization instruments and environmental
conditions. At the infrastructure level, SDLs will develop toward cloud-native, networked intelligent research
ecosystems. By coordinating distributed robot resources, AI agents can enable decentralized laboratories to
share knowledge and experience, thereby promoting unprecedented collaboration and innovation in material
science.

Although the prospects are broad, transforming this vision into reality still requires overcoming multiple
challenges. First, at the data foundation level, ensuring the quality, standardization, and interoperability of
material data from different laboratories, across scales, and multiple modalities is an important prerequisite
for collaborative innovation. Second, in terms of model capabilities, for material research and development
tasks involving critical decision making, there is an urgent need to enhance the explainability, reliability, and
robustness of AI foundation models to support high-confidence scientific judgments. Particularly critical are
challenges in ethics and governance. These include effectively avoiding algorithmic bias and ensuring new
materials meet ethical standards for safety and environmental impact, establishing clear frameworks for
ownership and responsibilities of independent research outcomes, and building robust data security and
intellectual property management mechanisms in the research ecosystems. Only by systematically addressing
these challenges can the field advance steadily in a responsible and sustainable direction.

With the development and integration of multimodal design, embodied intelligence, and cloud laboratories,
the establishment of next-generation research infrastructure centered on AI agents promises to accelerate
material science into an intelligent epoch. With continued evolution and deeper application in materials
discovery, we are expected to realize the vision of “on-demand material design”, providing solid support for
the innovation of high-performance materials in key fields such as energy, healthcare, and manufacturing,
and opening a new chapter in intelligent materials innovation.

DECLARATIONS
Authors’ contributions
Conception and design of the review: Ran, N.; Liu, J.