Intelligent Agent-Driven Design of Dynamically Reconstructed Metal-Organic Frameworks toward Industrial-Scale Anion-Exchange-Membrane Water Electrolysis

Abstract

Metal-organic frameworks (MOFs) are omnipotent precatalysts, yet their catalytically operative structures are typically reconstructed in situ under working conditions. Such dynamic evolution obscures the true active phase and renders conventional descriptor engineering and AI screening based on pristine MOFs intrinsically unreliable. Here, we develop a reconstruction-involved intelligent agent-driven discovery framework to integrate literature knowledge extraction, operando-informed active-phase identification, closed-loop density functional theory, and machine learning to design MOF-derived catalysts for the oxygen evolution reaction (OER) from their actually reconstructed states, rather than their precursor structures. By integrating a multimodal in situ and operando diagnostic platform, we uncover a programmable phase evolution from the pristine MOF to metal hydroxides and subsequently to the catalytically active oxyhydroxide phases, thereby establishing the structural basis for reconstruction-governed catalyst discovery. Applied to NiM-MOF systems, the framework identifies oxygen-vacancy-rich NiM-OOH as the operative catalytic motif and distills a transferable, physically interpretable descriptor that links Ni electronic structure and dopant properties to OER overpotential across reconstructed compositions. Guided by this descriptor, the workflow pinpoints NiFe-OOH/Ov derived from NiFe-MOF as the optimal anode, achieving 1.0 A cm^–2 at 1.88 V in an anion-exchange-membrane water electrolyzer and sustaining stable operation for 1100 h at 1.0 A cm^–2 and 500 h at 2.0 A cm^–2. Beyond identifying a high-performance NiFe system, this work provides a general blueprint for mechanistically grounded autonomous catalyst discovery in reconstruction-dominated electrocatalysis.