PILOT在环：长时程智能体的实时自我改进

摘要

长时程智能体运行会产生经验，这些经验既能改进当前运行，也能改善未来的工作。大多数自我改进方法只在执行结束后才处理这些经验，因此它们无法重定向正在进行的运行，也无法立即应用并验证从中获得的经验教训。我们认为，自我改进应当改为实时进行，既利用不断涌现的经验来重定向当前运行，又利用其更新持久化的框架。现有的智能体架构并不能完全支持这一目标。单智能体自我纠正在同一上下文中将任务执行与轨迹评估结合，而子智能体委派虽然将执行分离，但通常无法重定向正在运行的子智能体。我们提出了 PILOT，一个用于实时自我改进的监督者-工作者框架，它通过两种耦合机制实现：（1）实时引导，使独立的监督者能够在执行过程中重定向或中止当前工作者；（2）实时自我进化，将执行过程中暴露的流程与失败模式提炼为可复用的技能和记忆。在两个冻结骨干模型和三个基准测试上，PILOT 在六种配置中的五种排名第一。在 Terminal-Bench 2.0 上，PILOT 的表现比对照框架高出最多 9.8 个百分点。在自我改进设置中，PILOT 在 GLM-5.1 上提升了 14.6 分，在 Kimi-K2.6 上提升了 12.4 分。平均输出 token 数分别下降了 42.9% 和 47.4%，而每百万输出 token 的成功评估次数分别上升了 110.3% 和 134.0%。

English

Long-horizon agent runs generate experience that can improve both the current run and future work. Most self-improvement methods process this experience only after execution ends, so they cannot redirect the active run or immediately apply and validate lessons learned from it. We argue that self-improvement should instead be live, using emerging experience both to redirect the active run and to update the persistent harness. Existing agent architectures do not fully support this goal. Single-agent self-correction combines task execution and trajectory assessment within one context, while subagent delegation separates execution but typically cannot redirect an active subagent. We present PILOT, a supervisor-worker harness for live self-improvement through two coupled mechanisms: (1) live steering lets a separate supervisor redirect or abort the active worker during execution; and (2) live self-evolution distils procedures and failure modes revealed during execution into reusable skills and memory. Across two frozen backbones and three benchmarks, PILOT ranks first in five of six configurations. On Terminal-Bench 2.0, PILOT outperforms counterpart harnesses by up to 9.8 percentage points. In the self-improvement setting, PILOT gains 14.6 points with GLM-5.1 and 12.4 points with Kimi-K2.6. Mean output tokens fall by 42.9% and 47.4%, while successful evaluations per million output tokens rise by 110.3% and 134.0%, respectively.