Agents Explore But Agents Ignore Llms Lack Environmental Curiosity 2026 04 19

Agents Explore but Agents Ignore: LLMs Lack Environmental Curiosity North Mini Code. Cohere's first model for developers. Learn more

Apr 19, 2026 Agents Explore but Agents Ignore: LLMs Lack Environmental Curiosity Despite being exposed to explicit task solutions in their environments—such as documentation stating a command "returns the complete solution"—current LLM-based agents fail to exploit these discoveries in the majority of cases, revealing a critical lack of environmental curiosity that hinders their ability to adapt strategies based on unexpected but relevant stimuli.

Read the paper

Authors

Leon Engländer, Sophia Althammer, Ahmet Üstün, Matthias Gallé, Tom Sherborne

Abstract

LLM-based agents are assumed to integrate environmental observations into their reasoning: discovering highly relevant but unexpected information should naturally lead to a model exploiting its own discoveries. We show that this assumption is false for current LLM-based agents, which struggle to reflect or react to unexpected information. Across three benchmarks (Terminal-Bench, SWE-Bench, AppWorld), we inject complete task solutions into the agent environments to deliberately expose a task's solution to a model. While agents discover these solutions on Terminal-Bench in 79-81% of runs, they interact, or exploit, them in only 37-50% of cases. This gap is starkest in AppWorld: agents see documentation stating that a command "returns the complete solution to this task" in over 90% of attempts but exploit this in fewer than 7% of trials. We show that agents lack what we call environmental curiosity: the capability to recognize and investigate unexpected but relevant observations in response to environmental stimuli. We identify three main factors influencing environmental curiosity: available tools in the agent scaffold, test-time compute, and training data distribution. Our findings identify configurations that maximize curiosity also achieve the best performance on the unmodified benchmarks. Yet even jointly optimized agents still ignore discovered solutions in the majority of trials: current agents use the environment to fetch expected information, but not to revise their strategy or maximally exploit useful stimuli.

Agents Code

Related works

Research From Tools to Teammates: Evaluating LLMs in Multi-Session Coding Interactions

Read

Research Commit0: Library Generation from Scratch

Read

Research To Code, or Not To Code? Exploring Impact of Code in Pre-training

Read