Agent Manufacturing: Foundation-Model Agents as First-Class Industrial Entities

A new manufacturing paradigm called Agent Manufacturing is emerging, defined by the transfer of high-level production coordination from humans to foundation-model agents, according to a paper posted to arXiv. [1] The paper argues that manufacturing has evolved through four prior paradigms: mechanization, electrification, programmable automation, and Smart Manufacturing. Each stage shifted a specific type of work from humans to machines, but one layer remained fundamentally human throughout — the coordinative cognition of production. This layer includes the interpretive, allocative, diagnostic, negotiative, and governance work performed by engineers, planners, and operational managers. [1][2] The authors contend a fifth transition is now underway in which this coordinative layer, rather than physical or routine-cognitive tasks, is what foundation-model-based autonomous agents primarily redistribute. [1][2] Under the proposed definition, a manufacturing system qualifies as Agent Manufacturing when its principal coordination mechanism is reasoning performed by foundation-model agents. These agents must be capable of interpreting open-ended goals, planning over long horizons, invoking tools and machines, and negotiating with other agents and humans. [1][2] The definition is intentionally narrower and more falsifiable than concepts found in existing literature on cognitive manufacturing or Industry 5.0. It also distinguishes the paradigm from classical multi-agent manufacturing systems, which operated autonomously only within closed protocol spaces. [1][2] The concept of enterprise modelling, which involves the abstract representation of an organization's structure, processes, information, and resources to improve performance, provides a backdrop for understanding the complexity of the coordinative work now being targeted for redistribution. [5] The paper's framework suggests that the reasoning and negotiation capabilities of foundation-model agents could be applied to the dynamic, interpretive tasks that enterprise models seek to formalize. [1][5] The work was submitted on 24 May 2026 to the arXiv preprint server. [1]