How reasoning at action time helps language agents track changing states
Computation, Not Memory: Chain-of-Thought and State Tracking in LLM Agents
Machine LearningArtificial IntelligenceComputation and Language
Summary
Sometimes language models need to keep track of changing information to make the right decisions. This paper looks at how a model can figure out which files are locked when locks change but the new lock states aren't directly told. The authors find that the model does better when it reasons fresh each time before acting rather than relying on storing earlier thoughts. They also discover that what the model says during reasoning really matters more than just remembering past reasoning. Overall, generating reasoning at the moment of action helps the model follow rules better than just holding onto memory.
What this means in practice
- •For software engineers: Improve AI agents managing file systems by generating reasoning before decisions rather than relying on memory of past reasoning.
- •For robotics developers: Enhance robots' decision making in environments with changing states by focusing on immediate reasoning during action choice.
Authors
Pedram Mouseli, Massieh Moayedi
Abstract
Language-model agents often reason before they act, and the value of monitoring that reasoning depends on how it relates to their actions. We ask how an agent's actions depend on its reasoning when the correct action depends on a changing state. In a small file-system task, background events announce which file's lock has been toggled but not its new value. The agent therefore has to reconstruct the current locks from its history before it can move files safely. With Qwen3-8B and Qwen3.5-9B, we separate two roles of reasoning: generating it before each action, and retaining earlier reasoning in the prompt. Most of the reduction in lock-violating commands comes from reasoning written at the moment of action. Retaining earlier reasoning adds little to rule-following, although it helps one model complete the task. The benefit depends on what the reasoning says. Filler text, reasoning written for another history and an accurate table of the state do not reproduce it. When we give the agent reasoning written for a history that differs in one lock value, the agent acts on that reasoning even against its own history. Without reasoning, linear probes recover the state from the model's activations about as well as a text classifier recovers it from the transcript. However, decoding accuracy does not predict which interventions change the agent's actions. In this setting, the benefit of reasoning for rule-following comes mainly from generating it at each step. The state that can be decoded without reasoning is a limited guide to how the agent acts.