mlsanityData to 14 Oct 2026

Adapting Context Compression for Long-Horizon Agents with Counterfactual Continuations

arXiv 2609.36526 ↗First posted 29 Sep 2026cs.LG cs.AI cs.CLcited by 0

On the map: Efficient LLM inference and serving → Efficient reasoning and agent inference

Abstract (arXiv)

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.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Efficient reasoning and agent inference, one of 12 directions in Efficient LLM inference and serving. Builds on 19 of its 49 references that are in the map. First posted on arXiv 29 Sep 2026.

Builds on

19 of 49 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025ACON: Optimizing Context Compression for Long-horizon LLM Agents1 Oct

Other references · 18

  1. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  2. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,276
  3. 2023WebArena: A Realistic Web Environment for Building Autonomous Agentscited by 2,006
  4. 2025τ2-Bench: Evaluating Conversational Agents in a Dual-Control Environmentcited by 627
  5. 2024Executable Code Actions Elicit Better LLM Agentscited by 564
  6. 2023LongLLMLingua: Accelerating and Enhancing LLMs in Long Context Scenarios via Prompt Compressioncited by 453
  7. 2024AppWorld: A Controllable World of Apps and People for Benchmarking Interactive Coding Agentscited by 401
  8. 2023Learning to Compress Prompts with Gist Tokenscited by 390
  9. 2023Adapting Language Models to Compress Contextscited by 356
  10. 2024LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compressioncited by 348
  11. 2025MEM1: Learning to Synergize Memory and Reasoning for Efficient Long-Horizon Agentscited by 238
  12. 2023In-context Autoencoder for Context Compression in a Large Language Modelcited by 199
6 more
  1. 2023Compressing Context to Enhance Inference Efficiency of Large Language Modelscited by 189
  2. 2025Which Agent Causes Task Failures and When? On Automated Failure Attribution of LLM Multi-Agent Systemscited by 185
  3. 2025ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarizationcited by 141
  4. 2024OfficeBench: Benchmarking Language Agents across Multiple Applications for Office Automationcited by 70
  5. 2025Scaling LLM Multi-turn RL with End-to-end Summarization-based Context Managementcited by 51
  6. 2024TACO-RL: Task Aware Prompt Compression Optimization with Reinforcement Learningcited by 22

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Toward Reliable Context Compression for Long-Horizon Agents: An Empirical Study of Execution Instability
  2. 2026Control Under Compression: Reliability Frontiers for Tool-Using Agents
  3. 2026FOCUS: Training-Free Decision-Preserving Context Compression for LLM Agents
  4. 2026StateComp: Learning When to Compress History in Long Horizon Agents
  5. 2026A Self-Evolving Framework for Efficient Terminal Agents via Observational Context Compression
  6. 2026What Does Context Compression Cost an Agent? Interaction Costs Unrevealed by Task-Completion Metrics
  7. 2026Beyond Token Savings: A Systematic Study of Context Compression in LLM Agents
  8. 2026Experience Compression Spectrum: Unifying Memory, Skills, and Rules in LLM Agents