Paired prompt adaptation improves long horizon agent reliability
Adapting Context Compression for Long-Horizon Agents with Counterfactual Continuations
Machine LearningArtificial IntelligenceComputation and Language
Summary
Long-horizon agents keep track of many past actions to make decisions, but this history is hard to manage. The researchers found that compressing this history can cause mistakes that affect how reliably agents perform tasks, even before failing completely. To fix this, they developed a method called PAIR that finds exactly where the compression causes problems and adjusts the summaries to improve the agent’s performance. This new method helps agents act almost as well as if they had no compressed history at all.
What this means in practice
- •For autonomous system developers: Improve reliability of agents handling long sequences by adapting compression prompts to isolate and fix errors.
- •For game ai programmers: Enhance in-game agent decision making over long play sessions by refining history compression to prevent performance drops.
Authors
Guanghui Min, Liang Wu, Mingjia Shi, Yinhan He, Mayank Darbari, Liangjie Hong, Chen Chen
Abstract
Long-horizon agents require context compression to manage growing interaction histories. Compression quality, however, is ultimately determined by downstream execution. Existing prompt-adaptation methods infer compression errors by comparing full-context and compressed trajectories. Such comparisons cannot isolate individual compressions and are confounded by agent stochasticity. We first find that compression degrades reliability before solvability. Using matched counterfactual continuations that compare execution from the same agent state with versus without compression, we further show that severe degradation concentrates at isolated compression events. Motivated by this finding, we propose PAIR (Prompt Adaptation using Interventional Rollouts) for adapting structured compression prompts. PAIR identifies individual compressions that degrade subsequent execution, diagnoses their effects, and revises the relevant sections of a fixed compression template. PAIR achieves the strongest cross-run reliability among compressed methods in every main benchmark-scope combination, consistently exceeding the competing prompt-adaptation baseline. Without modifying the downstream agent, PAIR brings compressed execution close to the no-compression baseline and sometimes numerically exceeds it.