Do Enterprise Systems Need Learned World Models? The Importance of Context to Infer Dynamics

2026-05-12

AI Summary

This work addresses the significant performance degradation of conventional world models in enterprise systems under deployment shifts, caused by tenant-specific and time-varying business logic. To overcome this limitation, the paper introduces the Enterprise Discovery Agent, which uniquely treats runtime configuration reading as a core mechanism for dynamic reasoning. Instead of relying on static representations learned offline, the agent captures environmental transition dynamics by parsing system configurations in real time. The approach integrates configuration parsing, symbolic reasoning, and offline reinforcement learning, and introduces CascadeBench—a novel benchmark for evaluating cascade-effect prediction capabilities. Experimental results demonstrate that the discovery agent substantially outperforms traditional world models in out-of-distribution transfer scenarios, exhibiting markedly enhanced robustness.

Abstract

World models enable agents to anticipate the effects of their actions by internalizing environment dynamics. In enterprise systems, however, these dynamics are often defined by tenant-specific business logic that varies across deployments and evolves over time, making models trained on historical transitions brittle under deployment shift. We ask a question the world-models literature has not addressed: when the rules can be read at inference time, does an agent still need to learn them? We argue, and demonstrate empirically, that in settings where transition dynamics are configurable and readable, runtime discovery complements offline training by grounding predictions in the active system instance. We propose enterprise discovery agents, which recover relevant transition dynamics at runtime by reading the system's configuration rather than relying solely on internalized representations. We introduce CascadeBench, a reasoning-focused benchmark for enterprise cascade prediction that adopts the evaluation methodology of World of Workflows on diverse synthetic environments, and use it together with deployment-shift evaluation to show that offline-trained world models can perform well in-distribution but degrade as dynamics change, whereas discovery-based agents are more robust under shift by grounding their predictions in the current instance. Our findings suggest that, in configurable enterprise environments, agents should not rely solely on fixed internalized dynamics, but should incorporate mechanisms for discovering relevant transition logic at runtime.

Problem

• enterprise systems
• world models
• deployment shift
• transition dynamics
• runtime discovery

Innovation

• world models
• enterprise systems
• runtime discovery
• deployment shift
• configurable dynamics