Post-Training on Office Work Improves Software Engineering: A Behavioral Account of Cross-Domain Transfer
Not on the map. Why?
Abstract (arXiv)
Long-horizon tasks require agents to maintain coherent state and goals across nested and branching work. We call this capability goal-directed execution (GDE): the repeated application of four behaviors, namely selecting goals, constructing task-relevant state, maintaining fidelity to higher-level objectives, and verifying completion against the environment. We hypothesize that long-horizon post-training strengthens these behaviors across domains. We test this by post-training Qwen3.5-122B-A10B on 363 Long-Horizon Multi-Tool Agent (LHMTA) tasks drawn from office workflows. The collection contained no software-engineering tasks, yet the model's pass@1 improved by 5.8 points on SWE-Bench Pro. Matched trajectory analysis shows gains in all four GDE behaviors in both office workflows and software repositories. Aggregate SWE-Bench Pro statistics showed related changes in information gathering, implementation, and verification. Together, the results support a behavioral interpretation in which long-horizon post-training changed how the model organized and applied knowledge across tasks, with effects extending beyond the training domain.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 26 of its 49 references that are in the map. First posted on arXiv 3 Aug 2026.
Builds on
26 of 49 references.
References · 26
- 2022Training language models to follow instructions with human feedbackcited by 15,712
- 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
- 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
- 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,064
- 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
- 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 3,155
- 2022Scaling Instruction-Finetuned Language Modelscited by 2,997
- 2022Least-to-Most Prompting Enables Complex Reasoning in Large Language Modelscited by 1,397
- 2023LIMA: Less Is More for Alignmentcited by 1,080
- 2023Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Modelscited by 551
- 2024LESS: Selecting Influential Data for Targeted Instruction Tuningcited by 514
- 2019Leveraging Procedural Generation to Benchmark Reinforcement Learningcited by 478
14 more
- 2023Cognitive Architectures for Language Agentscited by 422
- 2025LLMs Get Lost In Multi-Turn Conversationcited by 400
- 2022Using cognitive psychology to understand GPT-3cited by 389
- 2022PlanBench: An Extensible Benchmark for Evaluating Large Language Models on Planning and Reasoning about Changecited by 372
- 2023What Makes Good Data for Alignment? A Comprehensive Study of Automatic Data Selection in Instruction Tuningcited by 354
- 2023ADaPT: As-Needed Decomposition and Planning with Language Modelscited by 172
- 2022Goal Misgeneralization: Why Correct Specifications Aren't Enough For Correct Goalscited by 120
- 2024HiAgent: Hierarchical Working Memory Management for Solving Long-Horizon Agent Tasks with Large Language Modelcited by 101
- 2025The Tool Decathlon: Benchmarking Language Agents for Diverse, Realistic, and Long-Horizon Task Executioncited by 73
- 2024Instruction Diversity Drives Generalization To Unseen Taskscited by 15
- 2025ReCAP: Recursive Context-Aware Reasoning and Planning for Large Language Model Agentscited by 13
- 2026EnterpriseBench Corecraft: Training Generalizable Agents on High-Fidelity RL Environmentscited by 8
- 2026Paying Less Generalization Tax: A Cross-Domain Generalization Study of RL Training for LLM Agentscited by 6
- 2026The Hierarchy of Agentic Capabilities: Evaluating Frontier Models on Realistic RL Environmentscited by 5
Built on by
Citing papers per year: 2026: 1.
1 paper.
- 2026Cross-Benchmark Transfer from RL on Agentic Coding Taskscited by 0
- 2026Cross-Benchmark Transfer from RL on Agentic Coding Tasks1 Oct 2026