Objectives and Key Results–Driven Multiagent Framework for Mechanical Design and Simulation

Abstract

Advanced structural and materials design increasingly depends on complex geometries and microstructures spanning continuum and atomistic descriptions. Yet the early idea–design–simulation loop remains manual: engineers translate informal intent into modeling assumptions, select physics, assemble solver inputs, and revise models as requirements evolve. Existing AI‐assisted design methods typically address well‐parameterized tasks within predefined spaces, offering limited support for this open‐ended exploratory phase. Here, we introduce an objectives and key results (OKR)‐agent, a coordinated peer multiagent framework that uses large language models (LLMs) to participate directly in mechanics design and simulation workflows. OKR‐agent organizes work around explicit OKR that encodes intent, task decomposition, and progress in a compact shared state. A supervisor, modeler, and simulator act as peers over an OKR board, coordinating reasoning with external continuum and atomistic solvers through a standardized tool set. This design keeps prompts subtask‐focused, limits context growth, and makes session evolution auditable. Across fiber‐bundle and polycrystalline‐metal molecular dynamics (MD) cases and plate‐with‐holes finite element method (FEM) benchmarks, the same team converts informal goals into executable simulations and interpretable outputs while accommodating mid‐session changes to geometry, boundary conditions, loading protocols, and microstructural architecture. A limitation remains: domain expertise is essential for validation and physical adequacy.