Researchers have proposed a new paradigm for simulating interactive environments: instead of rebuilding an executable system, let an agent serve as the environment itself. The paper, titled From Traces to Agentic Worlds: Agentic Language World Models for Interactive Environment Simulation, explores what the authors call agentic language world modeling.
The framework, Trace2Env, is described as learning-free and designed for situations where the original system is unavailable but historical interaction traces remain accessible. During an offline reconstruction phase, these traces are organized into a reusable environment "worldbook" containing environment schemas, transition rules, constraints, invariants, grounded evidence, and demonstrations.
At runtime, the world model agent actively consults the worldbook along with persistent episodic state and episodic memory to infer each action's observation and the state changes that should persist. A shared runtime harness verifies the proposal and commits accepted changes, allowing the consequences of earlier actions to carry into later turns.
According to the paper, Trace2Env was evaluated across nine environments, covering terminals, software repositories, Android and web apps, enterprise services, and text games. The authors report that it improves both next-observation fidelity and long-horizon interaction consistency compared with conventional prompt-based language world models.
In multi-turn interaction, the paper states that task agent actions generated against Trace2Env remain valid more often when replayed in the real environment, suggesting the simulated dynamics better preserve the effects of earlier actions across successive turns. The authors position agentic language world modeling as an alternative direction for building realistic environment replicas without reconstructing the original executable system.