Researchers propose Goal-Identity-Configurator architecture to distinguish agentive from agentic systems

A new arXiv pre-print titled “Critique of Agent Model” proposes a conceptual and architectural distinction between agentic and agentive systems, arguing that current LLM-based “agents” marketed as coding assistants or AI co-scientists typically lack genuine agency because their competence relies on external scaffolding rather than internalized structures.

The authors—Eric Xing, Mingkai Deng, and Jinyu Hou—ground their analysis in a philosophical tradition that ties agency to independent thought and survey contemporary agent architectures across five dimensions: goal, identity, decision-making, self-regulation, and learning.

They contend that genuine agency requires these structures to be internalized within the system itself, contrasting agentic systems—whose capabilities derive from engineered workflows—with agentive systems, whose capabilities, including social interaction, arise endogenously.

Building on this analysis, the paper introduces the Goal-Identity-Configurator (GIC) architecture for a general-purpose agent model, combining hierarchical goal decomposition, identity evolution, simulative reasoning grounded in a separately trained world model, learned self-regulation, and self-directed learning from both real and simulated experience.

The authors also discuss auditability, controllability, and safety for agentive systems that operate with greater autonomy while remaining under human oversight, framing their proposal within ongoing debates about existential risks from machine agency.